Add af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-2405e64f-1eba-4736-8f9f-c0fa4e22086f1760344445810-2025_10_13-10.34.47.298/source.csv

@@ -0,0 +1,25 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,5,"slurm/jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --job-name=train_dynamics_maskgit_grain_ablation\n#SBATCH --requeue\n#SBATCH --signal=b:usr1@300 # 5 min before timeout\n\n# --- signal trap to requeue job before timeout ---\nrequeue_job() {\n    echo ""[$(date)] caught sigusr1 (timeout warning), requeueing slurm job $SLURM_JOB_ID...""\n    # optional: trigger checkpoint saving here\n    # e.g., touch $checkpoint_dir/requeue_trigger\n    scontrol requeue $SLURM_JOB_ID\n    exit 0\n}\n\ntrap requeue_job sigusr1\n\n# set checkpoint flag based on restart count\nrestart_count=$(scontrol show job $SLURM_JOB_ID | grep -o 'Restarts=[0-9]*' | cut -d'=' -f2)\n\nif [ $restart_count -eq 0 ]; then\n    restore_ckpt_flag=""--no-restore-ckpt""\nelse\n    restore_ckpt_flag=""--restore-ckpt""\nfi\n\nexport CUDA_VISIBLE_DEVICES=0\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/npy_test\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/coinrun/maskgit/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_ckpt_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/train_tokenizer_default/3528955\n\nenv | grep SLURM\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    $restore_ckpt_flag \\n    --wandb_id $SLURM_JOB_ID \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=2048 \\n    --patch_size=16 \\n    --warmup_steps=0 \\n    --wsd_decay_steps=0 \\n    --num_steps=10_000 \\n    --log_image_interval=100_000 \\n    --log \\n    --log_checkpoint_interval=100_000 \\n    --name=coinrun-dynamics-maskgit-grain-ablation-bs2048-$slurm_job_id \\n    --tags coinrun dynamics maskgit grain-ablation bs2048 \\n    --entity instant-uv \\n    --project jafar \\n    --tokenizer_checkpoint=$tokenizer_ckpt_dir \\n    --data_dir $array_records_dir_train &\n\nchild_pid=$!\n\nwait $child_pid",shellscript,tab
+2,1626,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"10:34:47 AM [info] Activating crowd-code\n10:34:47 AM [info] Recording started\n10:34:47 AM [info] Initializing git provider using file system watchers...\n10:34:47 AM [info] Git repository found\n10:34:47 AM [info] Git provider initialized successfully\n10:34:48 AM [info] Initial git state: [object Object]\n",Log,tab
+3,60415,"slurm/jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh",0,0,"",shellscript,tab
+4,64045,"TERMINAL",0,0,"git status",,terminal_command
+5,64094,"TERMINAL",0,0,"]633;C",,terminal_output
+6,64345,"TERMINAL",0,0,"On branch prepend-action-maskgit\r\nYour branch is up to date with 'origin/prepend-action-maskgit'.\r\n\r\nUntracked files:\r\n  (use ""git add <file>..."" to include in what will be committed)\r\n\t checklist.md\r\n\tali-old-branch.diff\r\n\tappendix_c_nodes_embeddings_noise.md\r\n\tappendix_c_nodes_video_noise.md\r\n\tdata/_vizdoom.ini\r\n\tdata/jasmine_data/ViZDoomPPO/\r\n\tdata/jasmine_data/_vizdoom/\r\n\tdata/uv.lock\r\n\tdataset_duplicates.ipynb\r\n\tdiff.diff\r\n\tdiff2.diff\r\n\tdoom_job_starter.sh\r\n\tgifs/\r\n\tinput_pipeline/\r\n\tjasmine/train_dynamics_appendix-c.py\r\n\tjasmine/train_dynamics_appendix-c_main.py\r\n\tjasmine/train_dynamics_full_prec.py\r\n\tjasmine/train_tokenizer_full_precision.py\r\n\tkiller.sh\r\n\tkiller_partition.sh\r\n\tlog.log\r\n\tmessage.md\r\n\toverfit_dir.zip\r\n\trequirements-franz.txt\r\n\tsamples/\r\n\tscripts_cremers/\r\n\tslurm/\r\n\ttest.py\r\n\ttrain_dynamics_causal_3558251.log\r\n\tutils/\r\n\tuv.lock\r\n\r\nnothing added to commit but untracked files present (use ""git add"" to track)\r\n]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+7,66883,"TERMINAL",0,0,"cd slurm/",,terminal_command
+8,67149,"TERMINAL",0,0,"ls",,terminal_command
+9,67159,"TERMINAL",0,0,"]633;Ccommon  dev  jobs  README.md  templates  utils\r\n]0;tum_cte0515@hkn1991:~/Projects/jasmine/slurm",,terminal_output
+10,96722,"TERMINAL",0,0,"git status",,terminal_command
+11,96774,"TERMINAL",0,0,"]633;C",,terminal_output
+12,98177,"TERMINAL",0,0,"Refresh index:  89% (868/974)\r",,terminal_output
+13,98345,"TERMINAL",0,0,"Refresh index: 100% (974/974)\rRefresh index: 100% (974/974), done.\r\n",,terminal_output
+14,98511,"TERMINAL",0,0,"On branch main\r\nYour branch is up to date with 'origin/main'.\r\n\r\nChanges not staged for commit:\r\n  (use ""git add <file>..."" to update what will be committed)\r\n  (use ""git restore <file>..."" to discard changes in working directory)\r\n\tmodified:   jobs/mihir/horeka/coinrun/ablations/train_dyn_default-no-noise-main.sh\r\n\tmodified:   jobs/mihir/horeka/coinrun/default_runs/train_dyn_default.sh\r\n\r\nUntracked files:\r\n  (use ""git add <file>..."" to include in what will be committed)\r\n\tjobs/mihir/horeka/coinrun/ablations/appendix-c.sh\r\n\tjobs/mihir/horeka/coinrun/ablations/train_dyn_default-no-noise-ac-prepend.sh\r\n\tjobs/mihir/horeka/coinrun/causal_sweep/\r\n\tjobs/mihir/horeka/coinrun/default_runs/train_dyn_causal.sh\r\n\tjobs/mihir/horeka/coinrun/speed_ablation/\r\n\tjobs/mihir/horeka/doom/\r\n\tjobs/mihir/horeka/preprocessing/doom/\r\n\tjobs/mihir/horeka/preprocessing/train_doom_agent.sh\r\n\r\nno changes added to commit (use ""git add"" and/or ""git commit -a"")\r\n]0;tum_cte0515@hkn1991:~/Projects/jasmine/slurm",,terminal_output
+15,106569,"TERMINAL",0,0,"git add jobs/mihir/",,terminal_command
+16,106616,"TERMINAL",0,0,"]633;C",,terminal_output
+17,107438,"TERMINAL",0,0,"]0;tum_cte0515@hkn1991:~/Projects/jasmine/slurm",,terminal_output
+18,127741,"TERMINAL",0,0,"git commit -m ""added more scripts""",,terminal_command
+19,127744,"TERMINAL",0,0,"]633;C",,terminal_output
+20,128322,"TERMINAL",0,0,"[main 702e5e1] added more scripts\r\n 48 files changed, 2471 insertions(+), 14 deletions(-)\r\n create mode 100644 jobs/mihir/horeka/coinrun/ablations/appendix-c.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/ablations/train_dyn_default-no-noise-ac-prepend.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/causal_sweep/sweeper.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/causal_sweep/train_dyn_causal.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/default_runs/train_dyn_causal.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_1024/train_dyn_default.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_1024/train_dyn_default_ffn_dim_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_1024/train_dyn_default_flash_attn_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_1024/train_dyn_default_mixed_precision_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_1024/train_dyn_grain_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_default.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_default_ffn_dim_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_default_flash_attn_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_default_mixed_precision_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution120x160/train_dyn_default.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution120x160/train_dyn_default_500k.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution120x160/train_lam_default_1node.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution120x160/train_tokenizer_default_1node.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution120x160/train_tokenizer_default_1node_patch_size_4.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution120x160/train_tokenizer_default_1node_requeue.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/sample.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_dyn_default.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_dyn_default_500k.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_dyn_default_bigger.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_dyn_default_patch_size_4.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_lam_default_1gpu.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_tokenizer_default_1gpu.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_tokenizer_default_1gpu_500k.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_tokenizer_default_1gpu_bigger.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_tokenizer_default_1gpu_higher_lr.sh\r\n create mode 100644 jobs/mihir/horeka/doom/resolution60x80/train_tokenizer_default_1gpu_patch_size_4.sh\r\n create mode 100644 jobs/mihir/horeka/doom/train_tokenizer_default_1gpu.sh\r\n create mode 100644 jobs/mihir/horeka/doom/train_tokenizer_default_1node.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked._sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_10m.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_10m_120x160.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_10m_60x80.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_1m.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_1m_120x160.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_1m_60x80.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_50m.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_50m_120x160.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_50m_60x80.sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/doom/doom_chunked_test._sh\r\n create mode 100644 jobs/mihir/horeka/preprocessing/train_doom_agent.sh\r\n]0;tum_cte0515@hkn1991:~/Projects/jasmine/slurm",,terminal_output
+21,2694147,"TERMINAL",0,0,"cd ..",,terminal_command
+22,2694314,"TERMINAL",0,0,"ls",,terminal_command
+23,2694366,"TERMINAL",0,0,"]633;C",,terminal_output
+24,2694606,"TERMINAL",0,0," ali-old-branch.diff                    diff.diff             killer_partition.sh   __pycache__              tests\r\n appendix_c_nodes_embeddings_noise.md   doom_job_starter.sh   killer.sh             pyproject.toml           train_dynamics_causal_3558251.log\r\n appendix_c_nodes_video_noise.md        frame-knoms.png       LICENSE               README.md                utils\r\n' checklist.md'                         frame.png             log.log               requirements-franz.txt   uv.lock\r\n data                                   frames                logs                  samples                  wandb\r\n dataset_duplicates.ipynb               gifs                  message.md            scripts_cremers\r\n debug                                  input_pipeline        models                slurm\r\n diff2.diff                             jasmine               overfit_dir.zip       test.py\r\n]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-386afd79-6aa8-4e11-939b-d2205ae947e71760884674230-2025_10_19-16.38.53.787/source.csv

@@ -0,0 +1,168 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,1877,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"4:38:53 PM [info] Activating crowd-code\n4:38:53 PM [info] Recording started\n4:38:53 PM [info] Initializing git provider using file system watchers...\n4:38:54 PM [info] Git repository found\n4:38:54 PM [info] Git provider initialized successfully\n4:38:54 PM [info] Initial git state: [object Object]\n",Log,tab
+3,90716,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:1\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --job-name=train_dyn_single_gpu\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/train\narray_records_dir_val=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/val\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/dyn/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_checkpoint=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/interactive/3583033\n\nenv | grep SLURM\n\nexport JAX_COMPILER_ENABLE_REMAT_PASS=False\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    --image_height=64 \\n    --image_width=64 \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=8 \\n    --dyna_dim=2048 \\n    --dyna_num_blocks=16 \\n    --dyna_num_heads=32 \\n    --dyna_ffn_dim=8192 \\n    --num_steps=5000 \\n    --wsd_decay_steps=0 \\n    --warmup_steps=0 \\n    --patch_size=2 \\n    --log \\n    --name=coinrun-dyn-big-model-single-gpu-bs-8-$slurm_job_id \\n    --tags dyn coinrun big-model single-gpu \\n    --entity instant-uv \\n    --project jafar \\n    --data_dir $array_records_dir_train \\n    --tokenizer_checkpoint $tokenizer_checkpoint\n",shellscript,tab
+4,92083,"TERMINAL",0,0,"watch",,terminal_focus
+5,93886,"TERMINAL",0,0,"slurm/dev/franz/berlin/atari/dynamics_from_continued_from_40k_tokenizer/dynamics.shslurm/dev/franz/berlin/atari/dynamics_from_continued_from_40k_tokenizer/spawn_dynamics.shslurm/dev/franz/berlin/atari/tokenizer_lr_3e-6/slurm/dev/franz/berlin/atari/tokenizer_lr_3e-6/tokenizer.shslurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_continue_from_40k_to_200k/slurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_continue_from_40k_to_200k/spawn_tokenizers.shslurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_continue_from_40k_to_200k/tokenizer.shslurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_fleuret_hparams/slurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_fleuret_hparams/spawn_tokenizers.shslurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_fleuret_hparams/tokenizer.shslurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_fleuret_hparams_patch_size_4/slurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_fleuret_hparams_patch_size_4/spawn_tokenizers.shslurm/dev/franz/berlin/atari/tokenizer_lr_3e-6_fleuret_hparams_patch_size_4/tokenizer.shslurm/jobs/alfred/berlin/workshop/slurm/jobs/alfred/berlin/workshop/case_study/case_study_vanilla_genie_10M/slurm/jobs/alfred/berlin/workshop/case_study/case_study_vanilla_genie_10M/dynamics_case_study_dataset_10M.sbatchslurm/jobs/alfred/berlin/workshop/case_study/case_study_vanilla_genie_10M/sample_10M.sbatchslurm/jobs/alfred/berlin/workshop/case_study/case_study_vanilla_genie_10M/sample_10M_action_prepend.sbatchslurm/jobs/alfred/berlin/workshop/case_study/case_study_vanilla_genie_10M/tokenizer.sbatchslurm/jobs/alfred/berlin/workshop/case_study/misc/case_study_vanilla_genie/debug/slurm/jobs/alfred/berlin/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_10M_genie_default.shslurm/jobs/alfred/berlin/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_10M_jasmine_default.shslurm/jobs/alfred/berlin/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_50M_genie_default.shslurm/jobs/alfred/berlin/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_50M_jasmine_default.shslurm/jobs/alfred/berlin/workshop/speed_ablation/slurm/jobs/alfred/berlin/workshop/speed_ablation/chunk_ablation_runs/slurm/jobs/alfred/berlin/workshop/speed_ablation/chunk_ablation_runs/dynamics_chunk_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/chunk_ablation_runs/dynamics_chunk_ablation_spawner_batch_2048.shslurm/jobs/alfred/berlin/workshop/speed_ablation/chunk_ablation_runs/dynamics_chunk_ablation_spawner_batch_36.shslurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_ablation_dataset/slurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_ablation_dataset/generate_data_10M_chunksize.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_ablation_dataset/generate_data_10M_chunksize_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_size_ablation_dataset/slurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_1.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_10.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_100.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_1000.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_10000.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/slurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_base_speed_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_base_speed_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_ffn_dim_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_ffn_dim_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_flash_attn_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_flash_attn_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_grain_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_grain_ablation_spwaner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_no_flash_attn_full_prec.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/max_out/coinrun_dynamics_no_flash_attn_full_prec_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_base/slurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_base/coinrun_dynamics_base_speed_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_base/coinrun_dynamics_base_speed_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_ffn/slurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_flash_attention/slurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_full_prec/slurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec.sbatchslurm/jobs/alfred/berlin/workshop/speed_ablation/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec_spawner.shslurm/jobs/alfred/berlin/workshop/throughput/slurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_base/slurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_base/coinrun_dynamics_base_speed_ablation.sbatchslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_base/coinrun_dynamics_base_speed_ablation_batch_36.sbatchslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_base/coinrun_dynamics_base_speed_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_ffn/slurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation.sbatchslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_flash_attention/slurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation.sbatchslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation_spawner.shslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_full_prec/slurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec.sbatchslurm/jobs/alfred/berlin/workshop/throughput/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec_spawner.shslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/slurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/lam_jafar_reproduction_dataset_50M_requeue.sbatchslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_no_print.sbatchslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue.sbatchslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue_debug.sbatchslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_requeue.sbatchslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/slurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue.sbatchslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/tokenizer_jafar_reproduction_dataset_50M_requeue.sbatchslurm/jobs/franz/berlin/doom/generate_dataset/slurm/jobs/franz/berlin/doom/generate_dataset/generate_doom_dataset_10m.shslurm/jobs/mihir/horeka/coinrun/slurm/jobs/mihir/horeka/coinrun/big-model/slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.shslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.shslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.shslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.shslurm/jobs/mihir/horeka/coinrun/big-model/train_tokenizer_path_size_1.shslurm/jobs/mihir/horeka/coinrun/big-model/train_tokenizer_path_size_2.shslurm/jobs/mihir/horeka/coinrun/default_runs/train_tokenizer_default.shsent 253,918 bytes  received 1,881 bytes  24,361.81 bytes/sectotal size is 28,954,949  speedup is 113.19(jasmine) [tum_cte0515@hkn1993 jasmine]$ runner(jasmine) [tum_cte0515@hkn1993 jasmine_jobs]$ sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.shSubmitted batch job 3583200(jasmine) [tum_cte0515@hkn1993 jasmine_jobs]$ sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.shSubmitted batch job 3583201(jasmine) [tum_cte0515@hkn1993 jasmine_jobs]$ sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.shSubmitted batch job 3583203(jasmine) [tum_cte0515@hkn1993 jasmine_jobs]$ sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.shSubmitted batch job 3583204(jasmine) [tum_cte0515@hkn1993 jasmine_jobs]$ queue",,terminal_command
+6,95994,"TERMINAL",0,0,"scancel --me",,terminal_command
+7,96048,"TERMINAL",0,0,"]633;C",,terminal_output
+8,96061,"TERMINAL",0,0,"]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+9,98861,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:1\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --job-name=train_dyn_no_flash_attn_single_gpu\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/train\narray_records_dir_val=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/val\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/dyn/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_checkpoint=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/interactive/3583033\n\nenv | grep SLURM\n\nexport JAX_COMPILER_ENABLE_REMAT_PASS=False\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    --image_height=64 \\n    --image_width=64 \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=8 \\n    --num_steps=500 \\n    --wsd_decay_steps=0 \\n    --warmup_steps=0 \\n    --dyna_dim=2048 \\n    --dyna_num_blocks=16 \\n    --dyna_num_heads=32 \\n    --dyna_ffn_dim=8192 \\n    --patch_size=2 \\n    --no-use-flash-attention \\n    --log \\n    --name=coinrun-dyn-big-model-no-flash-attn-single-gpu-bs-8-$slurm_job_id \\n    --tags dyn coinrun big-model no-flash-attn single-gpu \\n    --entity instant-uv \\n    --project jafar \\n    --data_dir $array_records_dir_train \\n    --tokenizer_checkpoint $tokenizer_checkpoint\n",shellscript,tab
+10,102168,"TERMINAL",0,0,"bash",,terminal_focus
+11,104116,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --job-name=train_dyn_no_flash_attn_single_node\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/train\narray_records_dir_val=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/val\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/dyn/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_checkpoint=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/interactive/3583033\n\nenv | grep SLURM\n\nexport CUDA_VISIBLE_DEVICES=0\nexport JAX_COMPILER_ENABLE_REMAT_PASS=False\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    --image_height=64 \\n    --image_width=64 \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=1 \\n    --num_steps=5000 \\n    --wsd_decay_steps=0 \\n    --warmup_steps=0 \\n    --dyna_dim=2048 \\n    --dyna_num_blocks=16 \\n    --dyna_num_heads=32 \\n    --dyna_ffn_dim=8192 \\n    --patch_size=2 \\n    --no-use-flash-attention \\n    --log \\n    --name=coinrun-dyn-big-model-no-flash-attn-single-node-$slurm_job_id \\n    --tags dyn coinrun big-model no-flash-attn single-node \\n    --entity instant-uv \\n    --project jafar \\n    --data_dir $array_records_dir_train \\n    --tokenizer_checkpoint $tokenizer_checkpoint\n",shellscript,tab
+12,105277,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1431,0,"",shellscript,selection_mouse
+13,106237,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1430,1,"",shellscript,content
+14,106574,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1430,0,"8",shellscript,content
+15,106578,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1431,0,"",shellscript,selection_keyboard
+16,108220,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/%x_%j.log\n#SBATCH --job-name=train_dyn_single_node\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/train\narray_records_dir_val=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/coinrun_episodes_500m_gt_actions_split/val\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/dyn/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_checkpoint=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/interactive/3583033\n\nenv | grep SLURM\n\nexport CUDA_VISIBLE_DEVICES=0\nexport JAX_COMPILER_ENABLE_REMAT_PASS=False\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    --image_height=64 \\n    --image_width=64 \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=1 \\n    --dyna_dim=2048 \\n    --dyna_num_blocks=16 \\n    --dyna_num_heads=32 \\n    --dyna_ffn_dim=8192 \\n    --num_steps=5000 \\n    --wsd_decay_steps=0 \\n    --warmup_steps=0 \\n    --patch_size=2 \\n    --log \\n    --name=coinrun-dyn-big-model-single-node-$slurm_job_id \\n    --tags dyn coinrun big-model single-node \\n    --entity instant-uv \\n    --project jafar \\n    --data_dir $array_records_dir_train \\n    --tokenizer_checkpoint $tokenizer_checkpoint\n",shellscript,tab
+17,109184,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1417,0,"",shellscript,selection_mouse
+18,110121,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1416,1,"",shellscript,content
+19,110366,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1416,0,"8",shellscript,content
+20,110367,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1417,0,"",shellscript,selection_keyboard
+21,119207,"TERMINAL",0,0,"salloc --time=02:00:00 --partition=accelerated-h100 --nodes=1 --gres=gpu:1  --cpus-per-task=8",,terminal_command
+22,119269,"TERMINAL",0,0,"]633;Csalloc: Pending job allocation 3583319\r\nsalloc: job 3583319 queued and waiting for resources\r\n",,terminal_output
+23,120843,"TERMINAL",0,0,"bash",,terminal_focus
+24,121790,"TERMINAL",0,0,"idling",,terminal_command
+25,121898,"TERMINAL",0,0,"]633;C[?1049h(B[?7hEvery 1.0s: sinfo_t_idlehkn1993.localdomain: Sun Oct 19 16:40:55 2025Partition dev_cpuonly:     12 nodes idle\rPartition cpuonly:    191 nodes idle\rPartition dev_accelerated:\t 1 nodes idle\rPartition accelerated:     55 nodes idle\rPartition dev_accelerated-h100    :\t 0 nodes idle\rPartition accelerated-h100:\t 1 nodes idle\rPartition large:\t 8 nodes idle\rPartition accelerated-h200:\t 5 nodes idle",,terminal_output
+26,122973,"TERMINAL",0,0,"6",,terminal_output
+27,123970,"TERMINAL",0,0,"7",,terminal_output
+28,124388,"TERMINAL",0,0,"salloc",,terminal_focus
+29,125013,"TERMINAL",0,0,"8",,terminal_output
+30,126158,"TERMINAL",0,0,"9",,terminal_output
+31,126623,"TERMINAL",0,0,"^Csalloc: Job allocation 3583319 has been revoked.\r\nsalloc: Job aborted due to signal\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine",,terminal_output
+32,127092,"TERMINAL",0,0,"1:00",,terminal_output
+33,128163,"TERMINAL",0,0,"1",,terminal_output
+34,129175,"TERMINAL",0,0,"2",,terminal_output
+35,130211,"TERMINAL",0,0,"3",,terminal_output
+36,131259,"TERMINAL",0,0,"4",,terminal_output
+37,132299,"TERMINAL",0,0,"5",,terminal_output
+38,133357,"TERMINAL",0,0,"7",,terminal_output
+39,134383,"TERMINAL",0,0,"8",,terminal_output
+40,135088,"TERMINAL",0,0,"salloc --time=01:00:00 --partition=accelerated-h100 --nodes=1 --gres=gpu:1  --cpus-per-task=8",,terminal_command
+41,135158,"TERMINAL",0,0,"]633;Csalloc: Pending job allocation 3583320\r\nsalloc: job 3583320 queued and waiting for resources\r\n",,terminal_output
+42,135419,"TERMINAL",0,0,"9",,terminal_output
+43,136466,"TERMINAL",0,0,"10",,terminal_output
+44,136814,"TERMINAL",0,0,"watch",,terminal_focus
+45,137504,"TERMINAL",0,0,"1",,terminal_output
+46,137744,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+47,141543,"TERMINAL",0,0,"sync-runner",,terminal_command
+48,141603,"TERMINAL",0,0,"]633;Csending incremental file list\r\n",,terminal_output
+49,143233,"TERMINAL",0,0,"slurm/dev/franz/berlin/atari/\r\nslurm/dev/franz/berlin/atari/sample/\r\nslurm/dev/franz/berlin/atari/sample/sample_atari.sh\r\nslurm/dev/franz/berlin/atari/sample/spawn_sampler.sh\r\nslurm/dev/franz/berlin/coinrun/sample/maskgit/\r\nslurm/dev/franz/berlin/coinrun/sample/maskgit/sample_mila_submission.sh\r\nslurm/dev/franz/berlin/coinrun/sample/maskgit/sample_mila_submission_case_study_action_prepend.sh\r\nslurm/dev/franz/berlin/coinrun/sample/maskgit/sample_mila_submission_case_study_vanilla.sh\r\nslurm/jobs/alfred/berlin/workshop/\r\nslurm/jobs/alfred/berlin/workshop/sampling/\r\nslurm/jobs/alfred/berlin/workshop/sampling/sample_jasmine_action_add.sh\r\nslurm/jobs/alfred/berlin/workshop/sampling/sample_jasmine_action_prepend.sh\r\nslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/\r\nslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/dynamics_jafar_reproduction_dataset_50M.sbatch\r\nslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/dynamics_jafar_reproduction_dataset_50M_no_print.sbatch\r\nslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/\r\nslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue.sbatch\r\nslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue_new_env.sbatch\r\nslurm/jobs/alfred/berlin/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/tokenizer_jafar_reproduction_dataset_50M_requeue_new_env.sbatch\r\nslurm/jobs/alfred/horeka/\r\nslurm/jobs/alfred/horeka/workshop/\r\nslurm/jobs/alfred/horeka/workshop/case_study/\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/dynamics_case_study_dataset_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/dynamics_case_study_dataset_10M_action_prepend.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/sample_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/sample_10M_action_prepend.sbatch\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/tokenizer.sbatch\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/debug/\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/debug/lam.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_10M/debug/tokenizer.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_240M_w_chunking/\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_240M_w_chunking/tokenizer_chunking_100.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_240M_w_chunking/tokenizer_chunking_10k.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_50M/\r\nslurm/jobs/alfred/horeka/workshop/case_study/case_study_vanilla_genie_50M/tokenizer_action_prepend.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/batch_scaling/\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/batch_scaling/dynamics_batch_scaling.sbatch\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/batch_scaling/spawner.sh\r\n",,terminal_output
+50,144099,"TERMINAL",0,0,"slurm/jobs/alfred/horeka/workshop/case_study/misc/case_study/\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study/tokenizer.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study/debug/\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study/debug/tokenizer.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/tokenizer.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/dynamics_10M_patch_16.sbatch\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/dynamics_10M_patch_16_prepend.sbatch\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_10M_genie_default.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_10M_jasmine_default.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_50M.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_50M_genie_default.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_50M_jasmine_default.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/lam_50M_wsd.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/case_study_vanilla_genie/debug/tokenizer.sh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/non_co_training/\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/non_co_training/lam_10Msh\r\nslurm/jobs/alfred/horeka/workshop/case_study/misc/non_co_training/lam_50M.sh\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/generate_data_10M_arr_rec_no_seeding.sbatch\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/generate_data_10M_npy_arr_rec.sbatch\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/generate_data_240M_chunks_per_file_100.sbatch\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/generate_data_240M_chunks_per_file_10k.sbatch\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/generate_data_240M_npy_arr_rec.sbatch\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/generate_data_250M_npy_arr_rec.sbatch\r\nslurm/jobs/alfred/horeka/workshop/generate_coinrun_data/generate_data_250M_npy_arr_rec_test_val.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunk_ablation_runs/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunk_ablation_runs/dynamics_chunk_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunk_ablation_runs/dynamics_chunk_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_ablation_dataset/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_ablation_dataset/generate_data_10M_chunksize.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_ablation_dataset/generate_data_10M_chunksize_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_size_ablation_dataset/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_1.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_10.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_100.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_1000.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/chunking_size_ablation_dataset/generate_data_10M_chunks_per_file_10000.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_base_speed_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_base_speed_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_ffn_dim_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_ffn_dim_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_flash_attn_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_flash_attn_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_grain_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_grain_ablation_spwaner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_no_flash_attn_full_prec.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/max_out/coinrun_dynamics_no_flash_attn_full_prec_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_base/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_base/coinrun_dynamics_base_speed_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_base/coinrun_dynamics_base_speed_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_ffn/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_flash_attention/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_full_prec/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec.sbatch\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_2048/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_2048/train_dyn_default.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_2048/train_dyn_default_ffn_dim_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_2048/train_dyn_default_flash_attn_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_2048/train_dyn_default_mixed_precision_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_2048/train_dyn_grain_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_36/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_36/train_dyn_default.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_36/train_dyn_default_ffn_dim_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_36/train_dyn_default_flash_attn_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_36/train_dyn_default_mixed_precision_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/batch_size_36/train_dyn_grain_ablation.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/chunk/\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/chunk/train_chunk_per_file_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/chunk/train_chunk_size_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/speed_ablation_horeka/chunk/train_dynamics.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/\r\nslurm/jobs/alfred/horeka/workshop/throughput/dataset_gen/\r\nslurm/jobs/alfred/horeka/workshop/throughput/dataset_gen/generate_data_10M_chunksize.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/dataset_gen/generate_data_10M_chunksize_base.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/dataset_gen/generate_data_10M_chunksize_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/throughput/dataset_gen/chunking_size/\r\nslurm/jobs/alfred/horeka/workshop/throughput/dataset_gen/chunking_size/generate_data_10M_chunks_per_file.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/dataset_gen/chunking_size/generate_data_10M_chunks_per_file_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_base/\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_base/coinrun_dynamics_base_speed_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_base/coinrun_dynamics_base_speed_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_ffn/\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_ffn/coinrun_dynamics_ffn_dim_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_flash_attention/\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_flash_attention/coinrun_dynamics_flash_attn_ablation_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_full_prec/\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec.sbatch\r\nslurm/jobs/alfred/horeka/workshop/throughput/speed_ablation_full_prec/coinrun_dynamics_no_flash_attn_full_prec_spawner.sh\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/lam_jafar_reproduction_dataset_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/lam_jafar_reproduction_dataset_10M_auto_requeue_from_start.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/lam_jafar_reproduction_dataset_10M_auto_requeue_from_start_test.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/tokenizer_jafar_reproduction_dataset_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/tokenizer_jafar_reproduction_dataset_10M_auto_requeue.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/tokenizer_jafar_reproduction_dataset_10M_auto_requeue_from_start.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/misc/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/misc/tokenizer_jafar_reproduction_dataset_10M_patch_size_4.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/requeue/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/requeue/lam_jafar_reproduction_dataset_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/requeue/lam_jafar_reproduction_dataset_10M_requeue_auto.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/requeue/tokenizer_jafar_reproduction_dataset_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_10M/requeue/tokenizer_jafar_reproduction_dataset_10M_patch_size_4.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_250M/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_250M/lam_jafar_reproduction_dataset_250M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_250M/tokenizer_jafar_reproduction_dataset_250M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_250M/tokenizer_jafar_reproduction_dataset_250M_patch_size_4.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/lam_jafar_reproduction_dataset_50M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/lam_jafar_reproduction_dataset_50M_no_print.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/lam_jafar_reproduction_dataset_50M_no_print_requeue.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/lam_jafar_reproduction_dataset_50M_requeue.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_no_print.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue_debug.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/tokenizer_jafar_reproduction_dataset_50M_requeue.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/misc/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/misc/tokenizer_jafar_reproduction_dataset_10M_patch_size_4.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue/lam_jafar_reproduction_dataset_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue/lam_jafar_reproduction_dataset_10M_requeue_auto.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue/tokenizer_jafar_reproduction_dataset_10M.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue/tokenizer_jafar_reproduction_dataset_10M_patch_size_4.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/tokenizer_jafar_reproduction_dataset_50M_no_print_requeue.sbatch\r\nslurm/jobs/alfred/horeka/workshop/time_till_convergence/jafar_reproduction_50M/requeue_auto/tokenizer_jafar_reproduction_dataset_50M_requeue.sbatch\r\nslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh\r\nslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh\r\nslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh\r\nslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh\r\n",,terminal_output
+51,144596,"TERMINAL",0,0,"\r\nsent 286,316 bytes  received 3,068 bytes  82,681.14 bytes/sec\r\ntotal size is 29,164,292  speedup is 100.78\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+52,146262,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh",0,0,"",shellscript,tab
+53,147965,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",0,0,"",shellscript,tab
+54,149134,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1729,0,"",shellscript,selection_mouse
+55,150348,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1730,0,"bs-8-",shellscript,content
+56,150351,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1735,0,"",shellscript,selection_command
+57,151310,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1811,0,"",shellscript,selection_command
+58,151598,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1810,0,"",shellscript,selection_command
+59,152456,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",1810,0,"bs8 ",shellscript,content
+60,153686,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",0,0,"",shellscript,tab
+61,155527,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1703,0,"",shellscript,selection_mouse
+62,156169,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1704,0,"bs8 ",shellscript,content
+63,156171,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1708,0,"",shellscript,selection_command
+64,156981,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",0,0,"",shellscript,tab
+65,158715,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1671,0,"bs-8-",shellscript,content
+66,158717,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1676,0,"",shellscript,selection_command
+67,158976,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1737,0,"bs8 ",shellscript,content
+68,158977,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",1741,0,"",shellscript,selection_command
+69,161328,"TERMINAL",0,0,"salloc",,terminal_focus
+70,162247,"TERMINAL",0,0,"bash",,terminal_focus
+71,166759,"TERMINAL",0,0,"sync-runner",,terminal_command
+72,166837,"TERMINAL",0,0,"]633;Csending incremental file list\r\n",,terminal_output
+73,167005,"TERMINAL",0,0,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh\r\nslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh\r\nslurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh\r\n\r\nsent 66,436 bytes  received 432 bytes  133,736.00 bytes/sec\r\ntotal size is 29,164,314  speedup is 436.15\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+74,172214,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",0,0,"",shellscript,tab
+75,174071,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",0,0,"",shellscript,tab
+76,175194,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",0,0,"",shellscript,tab
+77,177851,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh",0,0,"",shellscript,tab
+78,179683,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh",1422,0,"",shellscript,selection_mouse
+79,181017,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh",1422,0,"0",shellscript,content
+80,181019,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh",1423,0,"",shellscript,selection_keyboard
+81,182276,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",0,0,"",shellscript,tab
+82,182910,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",0,0,"",shellscript,tab
+83,184987,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",0,0,"",shellscript,tab
+84,187194,"TERMINAL",0,0,"sync-runner",,terminal_command
+85,187247,"TERMINAL",0,0,"]633;Csending incremental file list\r\n",,terminal_output
+86,187409,"TERMINAL",0,0,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh\r\n\r\nsent 62,582 bytes  received 394 bytes  41,984.00 bytes/sec\r\ntotal size is 29,164,315  speedup is 463.10\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+87,191995,"TERMINAL",0,0,"sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh",,terminal_command
+88,192012,"TERMINAL",0,0,"]633;CSubmitted batch job 3583332\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+89,198751,"TERMINAL",0,0,"sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh",,terminal_command
+90,198801,"TERMINAL",0,0,"]633;C",,terminal_output
+91,198813,"TERMINAL",0,0,"Submitted batch job 3583333\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+92,202827,"TERMINAL",0,0,"sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh",,terminal_command
+93,202874,"TERMINAL",0,0,"]633;CSubmitted batch job 3583334\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+94,209916,"TERMINAL",0,0,"sbatch slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",,terminal_command
+95,209957,"TERMINAL",0,0,"]633;CSubmitted batch job 3583335\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+96,211074,"TERMINAL",0,0,"queue",,terminal_command
+97,211122,"TERMINAL",0,0,"]633;C",,terminal_output
+98,211181,"TERMINAL",0,0,"[?1049h(B[?7hEvery 1.0s: squeue --mehkn1993.localdomain: Sun Oct 19 16:42:24 2025JOBID PARTITION     NAME     USER ST\tTIME  NODES NODELIST(REASON)3583320 accelerat interact tum_cte0 PD\t0:00\t  1 (Priority)3583335 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)3583334 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)3583333 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)3583332 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)",,terminal_output
+99,212208,"TERMINAL",0,0,"5",,terminal_output
+100,213356,"TERMINAL",0,0,"6",,terminal_output
+101,214401,"TERMINAL",0,0,"8",,terminal_output
+102,215453,"TERMINAL",0,0,"9",,terminal_output
+103,215550,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_cte0515@hkn1993:~/Projects/jasmine_jobs",,terminal_output
+104,251258,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",0,0,"",shellscript,tab
+105,255449,"TERMINAL",0,0,"dev",,terminal_command
+106,256839,"TERMINAL",0,0,"cd slurm/",,terminal_command
+107,259312,"TERMINAL",0,0,"git status",,terminal_command
+108,259357,"TERMINAL",0,0,"]633;C",,terminal_output
+109,259791,"TERMINAL",0,0,"On branch main\r\nYour branch is up to date with 'origin/main'.\r\n\r\nChanges not staged for commit:\r\n  (use ""git add <file>..."" to update what will be committed)\r\n  (use ""git restore <file>..."" to discard changes in working directory)\r\n\tmodified:   jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model.sh\r\n\tmodified:   jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn.sh\r\n\tmodified:   jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_no_flash_attn_single_gpu.sh\r\n\tmodified:   jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh\r\n\r\nno changes added to commit (use ""git add"" and/or ""git commit -a"")\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/slurm",,terminal_output
+110,279284,"TERMINAL",0,0,"git add jobs/mihir/horeka/coinrun/big-model/",,terminal_command
+111,279342,"TERMINAL",0,0,"]633;C",,terminal_output
+112,279455,"TERMINAL",0,0,"]0;tum_cte0515@hkn1993:~/Projects/jasmine/slurm",,terminal_output
+113,307795,"TERMINAL",0,0,"git commit -m 'changed big model flash attn ablation to bs 8'",,terminal_command
+114,307857,"TERMINAL",0,0,"]633;C",,terminal_output
+115,308001,"TERMINAL",0,0,"[main 10fe9f3] changed big model flash attn ablation to bs 8\r\n 4 files changed, 13 insertions(+), 13 deletions(-)\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/slurm",,terminal_output
+116,309154,"TERMINAL",0,0,"git push",,terminal_command
+117,309210,"TERMINAL",0,0,"]633;C",,terminal_output
+118,310514,"TERMINAL",0,0,"Enumerating objects: 21, done.\r\nCounting objects:   4% (1/21)\rCounting objects:   9% (2/21)\rCounting objects:  14% (3/21)\rCounting objects:  19% (4/21)\rCounting objects:  23% (5/21)\rCounting objects:  28% (6/21)\rCounting objects:  33% (7/21)\rCounting objects:  38% (8/21)\rCounting objects:  42% (9/21)\rCounting objects:  47% (10/21)\rCounting objects:  52% (11/21)\rCounting objects:  57% (12/21)\rCounting objects:  61% (13/21)\rCounting objects:  66% (14/21)\rCounting objects:  71% (15/21)\rCounting objects:  76% (16/21)\rCounting objects:  80% (17/21)\rCounting objects:  85% (18/21)\rCounting objects:  90% (19/21)\rCounting objects:  95% (20/21)\rCounting objects: 100% (21/21)\rCounting objects: 100% (21/21), done.\r\nDelta compression using up to 152 threads\r\nCompressing objects:   9% (1/11)\rCompressing objects:  18% (2/11)\rCompressing objects:  27% (3/11)\rCompressing objects:  36% (4/11)\rCompressing objects:  45% (5/11)\rCompressing objects:  54% (6/11)\rCompressing objects:  63% (7/11)\rCompressing objects:  72% (8/11)\rCompressing objects:  81% (9/11)\rCompressing objects:  90% (10/11)\rCompressing objects: 100% (11/11)\rCompressing objects: 100% (11/11), done.\r\nWriting objects:   9% (1/11)\rWriting objects:  18% (2/11)\rWriting objects:  27% (3/11)\rWriting objects:  36% (4/11)\rWriting objects:  45% (5/11)\rWriting objects:  54% (6/11)\rWriting objects:  63% (7/11)\rWriting objects:  72% (8/11)\rWriting objects:  81% (9/11)\rWriting objects:  90% (10/11)\rWriting objects: 100% (11/11)\rWriting objects: 100% (11/11), 945 bytes | 945.00 KiB/s, done.\r\nTotal 11 (delta 9), reused 0 (delta 0), pack-reused 0\r\n",,terminal_output
+119,310646,"TERMINAL",0,0,"remote: Resolving deltas:   0% (0/9)\rremote: Resolving deltas:  11% (1/9)\rremote: Resolving deltas:  22% (2/9)\rremote: Resolving deltas:  33% (3/9)\rremote: Resolving deltas:  44% (4/9)\rremote: Resolving deltas:  55% (5/9)\rremote: Resolving deltas:  66% (6/9)\rremote: Resolving deltas:  77% (7/9)\rremote: Resolving deltas:  88% (8/9)\rremote: Resolving deltas: 100% (9/9)\rremote: Resolving deltas: 100% (9/9), completed with 9 local objects.\r\n",,terminal_output
+120,310695,"TERMINAL",0,0,"To github.com:p-doom/slurm.git\r\n   067a1f1..10fe9f3  main -> main\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/slurm",,terminal_output
+121,370586,"TERMINAL",0,0,"queue",,terminal_command
+122,370660,"TERMINAL",0,0,"]633;C[?1049h(B[?7hEvery 1.0s: squeue --mehkn1993.localdomain: Sun Oct 19 16:45:04 2025JOBID PARTITION     NAME     USER ST\tTIME  NODES NODELIST(REASON)3583333 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)3583334 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)3583335 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)3583332 accelerat train_dy tum_cte0 PD\t0:00\t  1 (Priority)3583320 accelerat interact tum_cte0 PD\t0:00\t  1 (Priority)",,terminal_output
+123,371716,"TERMINAL",0,0,"5",,terminal_output
+124,372751,"TERMINAL",0,0,"6",,terminal_output
+125,373798,"TERMINAL",0,0,"7",,terminal_output
+126,374848,"TERMINAL",0,0,"8",,terminal_output
+127,375883,"TERMINAL",0,0,"9",,terminal_output
+128,376937,"TERMINAL",0,0,"10",,terminal_output
+129,377972,"TERMINAL",0,0,"1",,terminal_output
+130,378819,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_cte0515@hkn1993:~/Projects/jasmine/slurm",,terminal_output
+131,383228,"TERMINAL",0,0,"cd $ws_dir",,terminal_command
+132,383955,"TERMINAL",0,0,"ls",,terminal_command
+133,384017,"TERMINAL",0,0,"]633;Ccheckpoints     data        data_breakout  data_doom       data_new     logs                                          scripts\r\ncount_items.sh  data_atari  data_coinrun   data_minecraft  huggingface  possibly_corrupt_files_in_this_workspace.txt\r\n]0;tum_cte0515@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared",,terminal_output
+134,389408,"TERMINAL",0,0,"cd data_doom/",,terminal_command
+135,389748,"TERMINAL",0,0,"ls",,terminal_command
+136,389801,"TERMINAL",0,0,"]633;C",,terminal_output
+137,389853,"TERMINAL",0,0,"dev                              doom_episodes_10m_bak           doom_episodes_1m_bak             doom_episodes_50m_bak\r\ndev_bak                          doom_episodes_10m_low_res_bak   doom_episodes_1m_low_res_bak     doom_episodes_50m_low_res_bak\r\ndoom_episodes_10m                doom_episodes_1m                doom_episodes_50m\r\ndoom_episodes_10m_120x160_fixed  doom_episodes_1m_120x160_fixed  doom_episodes_50m_120x160_fixed\r\ndoom_episodes_10m_60x80_fixed    doom_episodes_1m_60x80          doom_episodes_50m_60x80_fixed\r\n]0;tum_cte0515@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom",,terminal_output
+138,399058,"TERMINAL",0,0,"cd doom_episodes_50m_60x80_fixed",,terminal_command
+139,399453,"TERMINAL",0,0,"ls",,terminal_command
+140,399461,"TERMINAL",0,0,"]633;Cmetadata.json  train\r\n]0;tum_cte0515@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom/doom_episodes_50m_60x80_fixed",,terminal_output
+141,429513,"TERMINAL",0,0,"rm -rf *_bak",,terminal_command
+142,429520,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom/doom_episodes_50m_60x80_fixed",,terminal_output
+143,431212,"TERMINAL",0,0,"ls",,terminal_command
+144,431216,"TERMINAL",0,0,"]633;Cmetadata.json  train\r\n]0;tum_cte0515@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom/doom_episodes_50m_60x80_fixed",,terminal_output
+145,433494,"TERMINAL",0,0,"cd ..",,terminal_command
+146,436362,"TERMINAL",0,0,"rm -rf *_bak",,terminal_command
+147,436410,"TERMINAL",0,0,"]633;C",,terminal_output
+148,440124,"TERMINAL",0,0,"^C",,terminal_output
+149,440160,"TERMINAL",0,0,"\r\n]0;tum_cte0515@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom",,terminal_output
+150,440768,"TERMINAL",0,0,"ls",,terminal_command
+151,440784,"TERMINAL",0,0,"]633;Cdev                              doom_episodes_10m_bak           doom_episodes_1m_60x80        doom_episodes_50m_120x160_fixed\r\ndoom_episodes_10m                doom_episodes_10m_low_res_bak   doom_episodes_1m_bak          doom_episodes_50m_60x80_fixed\r\ndoom_episodes_10m_120x160_fixed  doom_episodes_1m                doom_episodes_1m_low_res_bak  doom_episodes_50m_bak\r\ndoom_episodes_10m_60x80_fixed    doom_episodes_1m_120x160_fixed  doom_episodes_50m             doom_episodes_50m_low_res_bak\r\n]0;tum_cte0515@hkn1993:/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom",,terminal_output
+152,447892,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1709,0,"",shellscript,selection_mouse
+153,447914,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1708,0,"",shellscript,selection_command
+154,448422,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1799,0,"",shellscript,selection_mouse
+155,448425,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1798,0,"",shellscript,selection_command
+156,448903,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1848,0,"",shellscript,selection_mouse
+157,448906,"slurm/jobs/mihir/horeka/coinrun/big-model/train_dyn_big_model_single_gpu.sh",1847,0,"",shellscript,selection_command
+158,562225,"jasmine/utils/nn.py",0,0,"import math\nfrom typing import Tuple, Callable, List\n\nfrom flax import nnx\nimport jax\nimport jax.numpy as jnp\nimport einops\n\n\ndef _get_spatiotemporal_positional_encoding(d_model: int, max_len: int = 5000):\n    """"""\n    Creates a function that applies separate sinusoidal positional encodings to the temporal and spatial dimensions.\n    """"""\n    pe = jnp.zeros((max_len, d_model))\n    position = jnp.arange(0, max_len, dtype=jnp.float32)[:, None]\n    div_term = jnp.exp(jnp.arange(0, d_model, 2) * (-math.log(10000.0) / d_model))\n    pe = pe.at[:, 0::2].set(jnp.sin(position * div_term))\n    pe = pe.at[:, 1::2].set(jnp.cos(position * div_term))\n\n    def _encode(x: jax.Array) -> jax.Array:\n        """"""\n        Args:\n            x: The input tensor of shape (Batch, Time, Space, Dimension).\n\n        Returns:\n            The input tensor with positional encodings added.\n        """"""\n        assert x.ndim == 4, f""Input must be 4-dimensional, but got shape {x.shape}""\n\n        num_timesteps = x.shape[1]\n        num_spatial_patches = x.shape[2]\n\n        # Temporal positional encoding: (1, T, 1, D)\n        temporal_pe = pe[None, :num_timesteps, None, :]\n        x = x + temporal_pe\n\n        # Spatial positional encoding: (1, 1, S, D)\n        spatial_pe = pe[None, None, :num_spatial_patches, :]\n        x = x + spatial_pe\n\n        return x\n\n    return _encode\n\n\nclass STBlock(nnx.Module):\n    def __init__(\n        self,\n        dim: int,\n        ffn_dim: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        rngs: nnx.Rngs,\n        sow_weights: bool,\n        sow_activations: bool,\n    ):\n        self.dim = dim\n        self.ffn_dim = ffn_dim\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.spatial_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.spatial_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.dim,\n            qkv_features=self.dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=False\n            ),\n            rngs=rngs,\n            decode=False,\n        )\n\n        self.temporal_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.temporal_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.dim,\n            qkv_features=self.dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                use_flash_attention=False, is_causal=True\n            ),\n            rngs=rngs,\n            decode=False,\n        )\n\n        self.ffn_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.ffn_dense1 = nnx.Linear(\n            in_features=self.dim,\n            out_features=self.ffn_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_dense2 = nnx.Linear(\n            in_features=self.ffn_dim,\n            out_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    @nnx.remat\n    def __call__(self, x_BTNM: jax.Array) -> jax.Array:\n        # --- Spatial attention ---\n        z_BTNM = self.spatial_norm(x_BTNM)\n        z_BTNM = self.spatial_attention(z_BTNM, sow_weights=self.sow_weights)\n        x_BTNM = x_BTNM + z_BTNM\n\n        # --- Temporal attention ---\n        x_BNTM = x_BTNM.swapaxes(1, 2)\n        z_BNTM = self.temporal_norm(x_BNTM)\n        z_BNTM = self.temporal_attention(z_BNTM, sow_weights=self.sow_weights)\n        x_BNTM = x_BNTM + z_BNTM\n        x_BTNM = x_BNTM.swapaxes(1, 2)\n\n        # --- Feedforward ---\n        z_BTNM = self.ffn_norm(x_BTNM)\n        z_BTND = self.ffn_dense1(z_BTNM)\n        z_BTND = jax.nn.gelu(z_BTND)\n        z_BTNM = self.ffn_dense2(z_BTND)\n        x_BTNM = x_BTNM + z_BTNM\n        if self.sow_activations:\n            self.sow(nnx.Intermediate, ""activations"", x_BTNM)\n        return x_BTNM\n\n\nclass STTransformer(nnx.Module):\n    """"""\n    Dimension keys:\n        B: batch size\n        T: number of frames\n        N: number of patches per frame\n        I: number of input features\n        M: model dimension\n        D: FFN dimension\n        V: vocabulary size\n    """"""\n\n    def __init__(\n        self,\n        input_dim: int,\n        model_dim: int,\n        ffn_dim: int,\n        out_dim: int,\n        num_blocks: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        rngs: nnx.Rngs,\n        sow_weights: bool = False,\n        sow_activations: bool = False,\n        sow_logits: bool = False,\n        max_len: int = 5000,\n    ):\n        self.input_dim = input_dim\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.out_dim = out_dim\n        self.num_blocks = num_blocks\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.sow_logits = sow_logits\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.input_norm1 = nnx.LayerNorm(\n            num_features=self.input_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.input_dense = nnx.Linear(\n            in_features=self.input_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_norm2 = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n\n        self.pos_enc = _get_spatiotemporal_positional_encoding(\n            self.model_dim, max_len=max_len\n        )\n\n        self.blocks = []\n        for _ in range(self.num_blocks):\n            self.blocks.append(\n                STBlock(\n                    dim=self.model_dim,\n                    ffn_dim=self.ffn_dim,\n                    num_heads=self.num_heads,\n                    dropout=self.dropout,\n                    param_dtype=self.param_dtype,\n                    dtype=self.dtype,\n                    use_flash_attention=self.use_flash_attention,\n                    rngs=rngs,\n                    sow_weights=self.sow_weights,\n                    sow_activations=self.sow_activations,\n                )\n            )\n\n        self.output_dense = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.out_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    def __call__(self, x_BTNI: jax.Array) -> jax.Array:\n        x_BTNI = self.input_norm1(x_BTNI)\n        x_BTNM = self.input_dense(x_BTNI)\n        x_BTNM = self.input_norm2(x_BTNM)\n        x_BTNM = self.pos_enc(x_BTNM)\n        for block in self.blocks:\n            x_BTNM = block(x_BTNM)\n\n        x_BTNV = self.output_dense(x_BTNM)\n        if self.sow_logits:\n            self.sow(nnx.Intermediate, ""logits"", x_BTNV)\n        return x_BTNV\n\n\nclass TransformerBlock(nnx.Module):\n    def __init__(\n        self,\n        model_dim: int,\n        ffn_dim: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        decode: bool,\n        rngs: nnx.Rngs,\n        sow_weights: bool,\n        sow_activations: bool,\n    ):\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.decode = decode\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.temporal_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.spatial_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.ffn_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.temporal_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.model_dim,\n            qkv_features=self.model_dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                use_flash_attention=False, is_causal=True\n            ),\n            rngs=rngs,\n            decode=self.decode,\n        )\n        self.spatial_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.model_dim,\n            qkv_features=self.model_dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=True\n            ),\n            rngs=rngs,\n            decode=self.decode,\n        )\n        self.ffn_dense1 = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.ffn_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_dense2 = nnx.Linear(\n            in_features=self.ffn_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    @nnx.remat\n    def __call__(\n        self, x_BTNM: jax.Array, pos_index: Tuple[jax.Array, jax.Array] | None = None\n    ) -> jax.Array:\n        # --- Spatial attention ---\n        B, T, N, M = x_BTNM.shape\n        z_FNM = einops.rearrange(x_BTNM, ""b t n m -> (b t) n m"")\n        z_FNM = self.spatial_norm(z_FNM)\n        z_FNM = self.spatial_attention(z_FNM, sow_weights=self.sow_weights)\n        z_BTNM = einops.rearrange(z_FNM, ""(b t) n m -> b t n m"", t=T)\n        x_BTNM = x_BTNM + z_BTNM\n        # --- Temporal attention ---\n        z_PTM = einops.rearrange(x_BTNM, ""b t n m -> (b n) t m"")\n        z_PTM = self.temporal_norm(z_PTM)\n        z_PTM = self.temporal_attention(z_PTM, sow_weights=self.sow_weights)\n        z_BTNM = einops.rearrange(z_PTM, ""(b n) t m -> b t n m"", n=N)\n        x_BTNM = x_BTNM + z_BTNM\n        # --- Feedforward ---\n        z_BTNM = self.ffn_norm(x_BTNM)\n        z_BTND = self.ffn_dense1(z_BTNM)\n        z_BTND = jax.nn.gelu(z_BTND)\n        z_BTNM = self.ffn_dense2(z_BTND)\n        x_BTNM = x_BTNM + z_BTNM\n        if self.sow_activations:\n            self.sow(nnx.Intermediate, ""activations"", x_BTNM)\n\n        return x_BTNM\n\n\nclass Transformer(nnx.Module):\n    """"""\n    Dimension keys:\n        B: batch size\n        T: number of frames\n        N: number of patches per frame\n        I: number of input features\n        M: model dimension\n        D: FFN dimension\n        V: vocabulary size\n        F: number of frames in batch\n        P: number of patch positions in batch\n    """"""\n\n    def __init__(\n        self,\n        input_dim: int,\n        model_dim: int,\n        ffn_dim: int,\n        out_dim: int,\n        num_blocks: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        decode: bool,\n        rngs: nnx.Rngs,\n        sow_logits: bool = False,\n        sow_weights: bool = False,\n        sow_activations: bool = False,\n        max_len: int = 5000,\n    ):\n        self.input_dim = input_dim\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.out_dim = out_dim\n        self.num_blocks = num_blocks\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.sow_logits = sow_logits\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.input_norm1 = nnx.LayerNorm(\n            num_features=self.input_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.input_dense = nnx.Linear(\n            in_features=self.input_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_norm2 = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n\n        self.pos_enc = _get_spatiotemporal_positional_encoding(\n            self.model_dim, max_len=max_len\n        )\n\n        self.blocks: List[TransformerBlock] = []\n        for _ in range(self.num_blocks):\n            self.blocks.append(\n                TransformerBlock(\n                    model_dim=self.model_dim,\n                    ffn_dim=self.ffn_dim,\n                    num_heads=self.num_heads,\n                    dropout=self.dropout,\n                    param_dtype=self.param_dtype,\n                    dtype=self.dtype,\n                    use_flash_attention=self.use_flash_attention,\n                    decode=decode,\n                    sow_weights=self.sow_weights,\n                    sow_activations=self.sow_activations,\n                    rngs=rngs,\n                )\n            )\n        self.output_dense = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.out_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    def __call__(\n        self, x_BTNI: jax.Array, pos_index: Tuple[jax.Array, jax.Array] | None = None\n    ) -> jax.Array:\n        x_BTNI = self.input_norm1(x_BTNI)\n        x_BTNM = self.input_dense(x_BTNI)\n        x_BTNM = self.input_norm2(x_BTNM)\n        x_BTNM = self.pos_enc(x_BTNM)\n        for block in self.blocks:\n            x_BTNM = block(x_BTNM, pos_index)\n\n        x_BTNV = self.output_dense(x_BTNM)\n        if self.sow_logits:\n            self.sow(nnx.Intermediate, ""logits"", x_BTNV)\n        return x_BTNV\n\n\ndef normalize(x: jax.Array) -> jax.Array:\n    return x / (jnp.linalg.norm(x, ord=2, axis=-1, keepdims=True) + 1e-8)\n\n\nclass VectorQuantizer(nnx.Module):\n    """"""\n    Dimension keys:\n        D: B * T * N\n        K: number of latents\n        L: latent dimension\n    """"""\n\n    def __init__(\n        self,\n        latent_dim: int,\n        num_latents: int,\n        dropout: float,\n        dtype: jnp.dtype,\n        rngs: nnx.Rngs,\n    ):\n        self.latent_dim = latent_dim\n        self.num_latents = num_latents\n        self.dropout = dropout\n        self.dtype = dtype\n\n        self.codebook = nnx.Param(\n            normalize(\n                nnx.initializers.normal(stddev=1)(\n                    rngs.params(), (self.num_latents, self.latent_dim)\n                )\n            )\n        )\n        self.drop = nnx.Dropout(self.dropout, rngs=rngs)\n\n    def __call__(\n        self, x_DL: jax.Array, training: bool\n    ) -> Tuple[jax.Array, jax.Array, jax.Array, jax.Array]:\n        # --- Compute distances ---\n        x_DL = x_DL.astype(self.dtype)\n        codebook = self.codebook.value.astype(self.dtype)\n\n        x_DL = normalize(x_DL)\n        normalized_codebook_KL = normalize(codebook)\n        distance_DK = -jnp.matmul(x_DL, normalized_codebook_KL.T)\n        if training:\n            distance_DK = self.drop(distance_DK)\n\n        # --- Get indices and embeddings ---\n        indices_D = jnp.argmin(distance_DK, axis=-1)\n        z_DL = codebook[indices_D]\n\n        # --- Straight through estimator ---\n        z_q_DL = x_DL + jax.lax.stop_gradient(z_DL - x_DL)\n        return z_q_DL, z_DL, x_DL, indices_D\n\n    def get_codes(self, indices_E: jax.Array) -> jax.Array:\n        return self.codebook[indices_E]\n\n\ndef _create_flash_attention_fn(use_flash_attention: bool, is_causal: bool) -> Callable:\n    """"""\n    Create an attention function that uses flash attention if enabled.\n\n    flax.nnx.MultiHeadAttention provides tensors with shape (batch..., length, num_heads, head_dim),\n    but jax.nn.dot_product_attention expects (batch, length, num_heads, head_dim). We reshape to\n    ensure compatibility. cuDNN's flash attention additionally requires a sequence length that\n    is a multiple of 4. We pad the sequence length to the nearest multiple of 4 and mask\n    accordingly. Note that cuDNN requires the mask to be broadcast before calling the attention\n    function due to strict shape checking.\n    """"""\n\n    def attention_fn(\n        query_BTHD, key_BSHD, value_BSHD, bias=None, mask_B111=None, **kwargs\n    ):\n        implementation = ""cudnn"" if use_flash_attention else None\n\n        def _merge_batch_dims(x):\n            return einops.rearrange(x, ""... l h k -> (...) l h k"")\n\n        def _pad(x, pad_size):\n            return jnp.pad(x, ((0, 0), (0, pad_size), (0, 0), (0, 0)))\n\n        original_shape = query_BTHD.shape\n        T = query_BTHD.shape[-3]\n        S = key_BSHD.shape[-3]\n\n        # Pad to nearest multiple of 4\n        Q = ((T + 3) // 4) * 4\n        pad_size_Q = Q - T\n        K = ((S + 3) // 4) * 4\n        pad_size_K = K - S\n\n        query_BQHD = _pad(_merge_batch_dims(query_BTHD), pad_size_Q)\n        key_BKHD = _pad(_merge_batch_dims(key_BSHD), pad_size_K)\n        value_BKHD = _pad(_merge_batch_dims(value_BSHD), pad_size_K)\n\n        attention_mask = jnp.ones((Q, K), dtype=jnp.bool_)\n        attention_mask = attention_mask.at[T:, :].set(False)\n        attention_mask = attention_mask.at[:, S:].set(False)\n\n        mask_11TS = attention_mask[jnp.newaxis, jnp.newaxis, :, :]\n\n        bias_4d = (\n            jnp.pad(\n                _merge_batch_dims(bias),\n                ((0, 0), (0, 0), (0, pad_size_Q), (0, pad_size_K)),\n            )\n            if bias is not None\n            else None\n        )\n\n        # NOTE: jax.nn.dot_product_attention does not support dropout\n        output_4d = jax.nn.dot_product_attention(\n            query=query_BQHD,\n            key=key_BKHD,\n            value=value_BKHD,\n            bias=bias_4d,\n            mask=mask_11TS,\n            implementation=implementation,\n            is_causal=is_causal,\n        )\n        return output_4d[..., :T, :, :].reshape(original_shape)\n\n    return attention_fn\n",python,tab
+159,575334,"jasmine/utils/preprocess.py",0,0,"import einops\nimport jax\nimport jax.numpy as jnp\n\n\ndef patchify(videos: jax.Array, size: int) -> jax.Array:\n    B, T, H, W, C = videos.shape\n    x = jnp.pad(videos, ((0, 0), (0, 0), (0, -H % size), (0, -W % size), (0, 0)))\n    return einops.rearrange(\n        x, ""b t (hn hp) (wn wp) c -> b t (hn wn) (hp wp c)"", hp=size, wp=size\n    )\n\n\ndef unpatchify(patches: jax.Array, size: int, h_out: int, w_out: int) -> jax.Array:\n    h_pad = -h_out % size\n    hn = (h_out + h_pad) // size\n    x = einops.rearrange(\n        patches,\n        ""b t (hn wn) (hp wp c) -> b t (hn hp) (wn wp) c"",\n        hp=size,\n        wp=size,\n        hn=hn,\n    )\n    return x[:, :, :h_out, :w_out]\n",python,tab
+160,579235,"jasmine/utils/preprocess.py",85,0,"",python,selection_mouse
+161,579320,"jasmine/utils/preprocess.py",83,4,"size",python,selection_mouse
+162,580825,"jasmine/utils/preprocess.py",86,0,"",python,selection_mouse
+163,604240,"jasmine/utils/nn.py",0,0,"",python,tab
+164,611046,"TERMINAL",0,0,"salloc",,terminal_focus
+165,763704,"TERMINAL",0,0,"salloc: job 3583320 has been allocated resources\r\nsalloc: Granted job allocation 3583320\r\n",,terminal_output
+166,763784,"TERMINAL",0,0,"salloc: Waiting for resource configuration\r\n",,terminal_output
+167,790873,"TERMINAL",0,0,"salloc: Nodes hkn0904 are ready for job\r\n",,terminal_output
+168,791597,"TERMINAL",0,0,"]0;tum_cte0515@hkn0904:~/Projects/jasmine[?2004h[tum_cte0515@hkn0904 jasmine]$ ",,terminal_output

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-59804d2c-df27-4b41-9fe6-ffdfef8825021760703869419-2025_10_17-14.25.51.972/source.csv

@@ -0,0 +1,3 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,3471,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"",Log,tab
+3,3682,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"2:25:50 PM [info] Activating crowd-code\n2:25:51 PM [info] Recording started\n2:25:52 PM [info] Initializing git provider using file system watchers...\n2:25:52 PM [info] No workspace folder found\n2:25:55 PM [info] Retrying git provider initialization...\n2:25:55 PM [info] No workspace folder found\n",Log,content

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-5b002f72-29b2-49c4-8fb4-15e9a0f2c68a1760283746731-2025_10_12-17.43.07.478/source.csv

@@ -0,0 +1,28 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,5,"slurm/jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --job-name=train_dynamics_maskgit_grain_ablation\n#SBATCH --requeue\n#SBATCH --signal=b:usr1@300 # 5 min before timeout\n\n# --- signal trap to requeue job before timeout ---\nrequeue_job() {\n    echo ""[$(date)] caught sigusr1 (timeout warning), requeueing slurm job $SLURM_JOB_ID...""\n    # optional: trigger checkpoint saving here\n    # e.g., touch $checkpoint_dir/requeue_trigger\n    scontrol requeue $SLURM_JOB_ID\n    exit 0\n}\n\ntrap requeue_job sigusr1\n\n# set checkpoint flag based on restart count\nrestart_count=$(scontrol show job $SLURM_JOB_ID | grep -o 'Restarts=[0-9]*' | cut -d'=' -f2)\n\nif [ $restart_count -eq 0 ]; then\n    restore_ckpt_flag=""--no-restore-ckpt""\nelse\n    restore_ckpt_flag=""--restore-ckpt""\nfi\n\nexport CUDA_VISIBLE_DEVICES=0\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/npy_test\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/coinrun/maskgit/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_ckpt_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/train_tokenizer_default/3528955\n\nenv | grep SLURM\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    $restore_ckpt_flag \\n    --wandb_id $SLURM_JOB_ID \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=2048 \\n    --patch_size=16 \\n    --warmup_steps=0 \\n    --wsd_decay_steps=0 \\n    --num_steps=10_000 \\n    --log_image_interval=100_000 \\n    --log \\n    --log_checkpoint_interval=100_000 \\n    --name=coinrun-dynamics-maskgit-grain-ablation-bs2048-$slurm_job_id \\n    --tags coinrun dynamics maskgit grain-ablation bs2048 \\n    --entity instant-uv \\n    --project jafar \\n    --tokenizer_checkpoint=$tokenizer_ckpt_dir \\n    --data_dir $array_records_dir_train &\n\nchild_pid=$!\n\nwait $child_pid",shellscript,tab
+2,1762,"TERMINAL",0,0,"",,terminal_focus
+3,2575,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"5:43:07 PM [info] Activating crowd-code\n5:43:07 PM [info] Recording started\n5:43:07 PM [info] Initializing git provider using file system watchers...\n5:43:08 PM [info] Git repository found\n5:43:08 PM [info] Git provider initialized successfully\n5:43:08 PM [info] Initial git state: [object Object]\n",Log,tab
+4,3422,"slurm/jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh",0,0,"",shellscript,tab
+5,4729,"TERMINAL",0,0,"source /home/hk-project-p0023960/tum_cte0515/Projects/jasmine/.venv/bin/activate",,terminal_command
+6,42983,"TERMINAL",0,0,"scancel 3562087",,terminal_command
+7,43003,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+8,57931,"TERMINAL",0,0,"scancel 3562088",,terminal_command
+9,57955,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+10,73562,"TERMINAL",0,0,"scancel 3562089",,terminal_command
+11,73582,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+12,87435,"TERMINAL",0,0,"scancel 3562081",,terminal_command
+13,87459,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+14,95439,"TERMINAL",0,0,"scancel 3562082",,terminal_command
+15,95459,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+16,105395,"TERMINAL",0,0,"scancel 3562083",,terminal_command
+17,105420,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+18,114078,"TERMINAL",0,0,"scancel 3562076",,terminal_command
+19,114105,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+20,119314,"TERMINAL",0,0,"scancel 3562077",,terminal_command
+21,119343,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+22,180469,"TERMINAL",0,0,"scancel 3562078",,terminal_command
+23,180518,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+24,199040,"TERMINAL",0,0,"scancel 3562084",,terminal_command
+25,199065,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output
+26,223683,"TERMINAL",0,0,"scancel 3562090",,terminal_command
+27,223700,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1991:~/Projects/jasmine",,terminal_output

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-b510dc45-0390-4ba1-9cd3-8518acd9730d1761060552425-2025_10_21-17.29.37.898/source.csv

@@ -0,0 +1,46 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,3909,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"5:29:37 PM [info] Activating crowd-code\n5:29:37 PM [info] Recording started\n5:29:37 PM [info] Initializing git provider using file system watchers...\n5:29:38 PM [info] Git repository found\n5:29:38 PM [info] Git provider initialized successfully\n5:29:38 PM [info] Initial git state: [object Object]\n",Log,tab
+3,6216,"TERMINAL",0,0,"",,terminal_command
+4,6232,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+5,13635,"TERMINAL",0,0,"",,terminal_command
+6,49183,"TERMINAL",0,0,"",,terminal_focus
+7,51836,"TERMINAL",0,0,"git branch",,terminal_command
+8,51916,"TERMINAL",0,0,"]633;C[?1h=\r* feat/diffusion-backend\r\n  main\r\n  vizdoom-dataset\r\n\r[?1l>]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+9,53098,"TERMINAL",0,0,"git pull",,terminal_command
+10,53149,"TERMINAL",0,0,"]633;C",,terminal_output
+11,54878,"TERMINAL",0,0,"remote: Enumerating objects: 18, done.\r\nremote: Counting objects:   5% (1/18)\rremote: Counting objects:  11% (2/18)\rremote: Counting objects:  16% (3/18)\rremote: Counting objects:  22% (4/18)\rremote: Counting objects:  27% (5/18)\rremote: Counting objects:  33% (6/18)\rremote: Counting objects:  38% (7/18)\rremote: Counting objects:  44% (8/18)\rremote: Counting objects:  50% (9/18)\rremote: Counting objects:  55% (10/18)\rremote: Counting objects:  61% (11/18)\rremote: Counting objects:  66% (12/18)\rremote: Counting objects:  72% (13/18)\rremote: Counting objects:  77% (14/18)\rremote: Counting objects:  83% (15/18)\rremote: Counting objects:  88% (16/18)\rremote: Counting objects:  94% (17/18)\rremote: Counting objects: 100% (18/18)\rremote: Counting objects: 100% (18/18), done.\r\nremote: Compressing objects:  14% (1/7)\rremote: Compressing objects:  28% (2/7)\rremote: Compressing objects:  42% (3/7)\rremote: Compressing objects:  57% (4/7)\rremote: Compressing objects:  71% (5/7)\rremote: Compressing objects:  85% (6/7)\rremote: Compressing objects: 100% (7/7)\rremote: Compressing objects: 100% (7/7), done.\r\nremote: Total 18 (delta 11), reused 18 (delta 11), pack-reused 0 (from 0)\r\nUnpacking objects:   5% (1/18)\rUnpacking objects:  11% (2/18)\rUnpacking objects:  16% (3/18)\rUnpacking objects:  22% (4/18)\rUnpacking objects:  27% (5/18)\rUnpacking objects:  33% (6/18)\rUnpacking objects:  38% (7/18)\rUnpacking objects:  44% (8/18)\rUnpacking objects:  50% (9/18)\rUnpacking objects:  55% (10/18)\rUnpacking objects:  61% (11/18)\rUnpacking objects:  66% (12/18)\rUnpacking objects:  72% (13/18)\rUnpacking objects:  77% (14/18)\rUnpacking objects:  83% (15/18)\rUnpacking objects:  88% (16/18)\rUnpacking objects:  94% (17/18)\rUnpacking objects: 100% (18/18)\rUnpacking objects: 100% (18/18), 9.42 KiB | 79.00 KiB/s, done.\r\n",,terminal_output
+12,54958,"TERMINAL",0,0,"From github.com:p-doom/jasmine\r\n * [new branch]      feat/diffusion-forcing -> origin/feat/diffusion-forcing\r\n",,terminal_output
+13,54985,"TERMINAL",0,0,"Already up to date.\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+14,58816,"TERMINAL",0,0,"git checkout feat/diffusion-forcing",,terminal_command
+15,58866,"TERMINAL",0,0,"]633;C",,terminal_output
+16,58956,"TERMINAL",0,0,"branch 'feat/diffusion-forcing' set up to track 'origin/feat/diffusion-forcing'.\r\nSwitched to a new branch 'feat/diffusion-forcing'\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+17,60163,"TERMINAL",0,0,"git pull",,terminal_command
+18,60198,"TERMINAL",0,0,"]633;C",,terminal_output
+19,60567,"",0,0,"Switched from branch 'feat/diffusion-backend' to 'feat/diffusion-forcing'",,git_branch_checkout
+20,60940,"TERMINAL",0,0,"c",,terminal_output
+21,61068,"TERMINAL",0,0,"d",,terminal_output
+22,61160,"TERMINAL",0,0," ",,terminal_output
+23,61827,"TERMINAL",0,0,"Already up to date.\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+24,63858,"TERMINAL",0,0,"cd slurm/",,terminal_command
+25,63877,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1/slurm",,terminal_output
+26,66596,"TERMINAL",0,0,"git pull",,terminal_command
+27,66631,"TERMINAL",0,0,"]633;C",,terminal_output
+28,68249,"TERMINAL",0,0,"remote: Enumerating objects: 13, done.\r\nremote: Counting objects:   7% (1/13)\rremote: Counting objects:  15% (2/13)\rremote: Counting objects:  23% (3/13)\rremote: Counting objects:  30% (4/13)\rremote: Counting objects:  38% (5/13)\rremote: Counting objects:  46% (6/13)\rremote: Counting objects:  53% (7/13)\rremote: Counting objects:  61% (8/13)\rremote: Counting objects:  69% (9/13)\rremote: Counting objects:  76% (10/13)\rremote: Counting objects:  84% (11/13)\rremote: Counting objects:  92% (12/13)\rremote: Counting objects: 100% (13/13)\rremote: Counting objects: 100% (13/13), done.\r\nremote: Compressing objects:  25% (1/4)\rremote: Compressing objects:  50% (2/4)\rremote: Compressing objects:  75% (3/4)\rremote: Compressing objects: 100% (4/4)\rremote: Compressing objects: 100% (4/4), done.\r\nremote: Total 7 (delta 3), reused 7 (delta 3), pack-reused 0 (from 0)\r\nUnpacking objects:  14% (1/7)\rUnpacking objects:  28% (2/7)\rUnpacking objects:  42% (3/7)\rUnpacking objects:  57% (4/7)\rUnpacking objects:  71% (5/7)\rUnpacking objects:  85% (6/7)\rUnpacking objects: 100% (7/7)\rUnpacking objects: 100% (7/7), 642 bytes | 13.00 KiB/s, done.\r\n",,terminal_output
+29,68357,"TERMINAL",0,0,"From github.com:p-doom/slurm\r\n   14e9cef..9f53682  main       -> origin/main\r\n",,terminal_output
+30,68422,"TERMINAL",0,0,"Updating 14e9cef..9f53682\r\n",,terminal_output
+31,68561,"TERMINAL",0,0,"Fast-forward\r\n jobs/mihir/berlin/diffusion/dyn_dit_breakout.sh | 3 ++-\r\n 1 file changed, 2 insertions(+), 1 deletion(-)\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1/slurm",,terminal_output
+32,103592,"TERMINAL",0,0,"",,terminal_command
+33,103604,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+34,160099,"TERMINAL",0,0,"git pull",,terminal_command
+35,160149,"TERMINAL",0,0,"]633;C",,terminal_output
+36,161751,"TERMINAL",0,0,"remote: Enumerating objects: 14, done.\r\nremote: Counting objects:   7% (1/14)\rremote: Counting objects:  14% (2/14)\rremote: Counting objects:  21% (3/14)\rremote: Counting objects:  28% (4/14)\rremote: Counting objects:  35% (5/14)\rremote: Counting objects:  42% (6/14)\rremote: Counting objects:  50% (7/14)\rremote: Counting objects:  57% (8/14)\rremote: Counting objects:  64% (9/14)\rremote: Counting objects:  71% (10/14)\rremote: Counting objects:  78% (11/14)\rremote: Counting objects:  85% (12/14)\rremote: Counting objects:  92% (13/14)\rremote: Counting objects: 100% (14/14)\rremote: Counting objects: 100% (14/14), done.\r\nremote: Compressing objects:  20% (1/5)\rremote: Compressing objects:  40% (2/5)\rremote: Compressing objects:  60% (3/5)\rremote: Compressing objects:  80% (4/5)\rremote: Compressing objects: 100% (5/5)\rremote: Compressing objects: 100% (5/5), done.\r\nremote: Total 8 (delta 3), reused 8 (delta 3), pack-reused 0 (from 0)\r\nUnpacking objects:  12% (1/8)\rUnpacking objects:  25% (2/8)\rUnpacking objects:  37% (3/8)\rUnpacking objects:  50% (4/8)\rUnpacking objects:  62% (5/8)\rUnpacking objects:  75% (6/8)\rUnpacking objects:  87% (7/8)\rUnpacking objects: 100% (8/8)\rUnpacking objects: 100% (8/8), 1.19 KiB | 25.00 KiB/s, done.\r\nFrom github.com:p-doom/slurm\r\n   9f53682..15cb084  main       -> origin/main\r\n",,terminal_output
+37,161815,"TERMINAL",0,0,"Updating 9f53682..15cb084\r\n",,terminal_output
+38,162012,"TERMINAL",0,0,"Fast-forward\r\n jobs/mihir/berlin/diffusion/dyn_dit_breakout_dev.sh | 46 ++++++++++++++++++++++++++++++++\r\n jobs/mihir/berlin/diffusion/dyn_dit_coinrun.sh      |  4 +--\r\n 2 files changed, 48 insertions(+), 2 deletions(-)\r\n create mode 100644 jobs/mihir/berlin/diffusion/dyn_dit_breakout_dev.sh\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1/slurm",,terminal_output
+39,164992,"TERMINAL",0,0,"cd ..",,terminal_command
+40,165020,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+41,175690,"TERMINAL",0,0,"source ../jasmine/.venv/bin/activate",,terminal_command
+42,175698,"TERMINAL",0,0,"]633;C]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+43,216697,"TERMINAL",0,0,"sbatch slurm/jobs/mihir/berlin/diffusion/dyn_dit_coinrun.sh",,terminal_command
+44,216728,"TERMINAL",0,0,"]633;CSubmitted batch job 32406\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output
+45,263827,"TERMINAL",0,0,"sbatch slurm/jobs/mihir/berlin/diffusion/dyn_dit_breakout.sh",,terminal_command
+46,263927,"TERMINAL",0,0,"]633;CSubmitted batch job 32407\r\n]0;mihir.mahajan@hai-login2:~/Projects/jasmine_runner_1",,terminal_output

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-e1782904-5e7c-4264-b342-b03f67dbe6421760704011771-2025_10_17-14.28.12.326/source.csv

@@ -0,0 +1,24 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,5,"slurm/jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --job-name=train_dynamics_maskgit_grain_ablation\n#SBATCH --requeue\n#SBATCH --signal=b:usr1@300 # 5 min before timeout\n\n# --- signal trap to requeue job before timeout ---\nrequeue_job() {\n    echo ""[$(date)] caught sigusr1 (timeout warning), requeueing slurm job $SLURM_JOB_ID...""\n    # optional: trigger checkpoint saving here\n    # e.g., touch $checkpoint_dir/requeue_trigger\n    scontrol requeue $SLURM_JOB_ID\n    exit 0\n}\n\ntrap requeue_job sigusr1\n\n# set checkpoint flag based on restart count\nrestart_count=$(scontrol show job $SLURM_JOB_ID | grep -o 'Restarts=[0-9]*' | cut -d'=' -f2)\n\nif [ $restart_count -eq 0 ]; then\n    restore_ckpt_flag=""--no-restore-ckpt""\nelse\n    restore_ckpt_flag=""--restore-ckpt""\nfi\n\nexport CUDA_VISIBLE_DEVICES=0\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/npy_test\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/coinrun/maskgit/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_ckpt_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/train_tokenizer_default/3528955\n\nenv | grep SLURM\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    $restore_ckpt_flag \\n    --wandb_id $SLURM_JOB_ID \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=2048 \\n    --patch_size=16 \\n    --warmup_steps=0 \\n    --wsd_decay_steps=0 \\n    --num_steps=10_000 \\n    --log_image_interval=100_000 \\n    --log \\n    --log_checkpoint_interval=100_000 \\n    --name=coinrun-dynamics-maskgit-grain-ablation-bs2048-$slurm_job_id \\n    --tags coinrun dynamics maskgit grain-ablation bs2048 \\n    --entity instant-uv \\n    --project jafar \\n    --tokenizer_checkpoint=$tokenizer_ckpt_dir \\n    --data_dir $array_records_dir_train &\n\nchild_pid=$!\n\nwait $child_pid",shellscript,tab
+2,1991,"anysphere.remote-ssh.Remote - SSH",0,0,"2025-10-17 14:27:11.423 [info] Resolving ssh remote authority 'horeka' (Unparsed 'ssh-remote+7b22686f73744e616d65223a22686f72656b61227d') (attempt #1)\n2025-10-17 14:27:12.383 [info] SSH askpass server listening on /var/folders/mr/xp86mpcd01sd72mft8vjq27r0000gn/T/cursor-ssh-Vf6muF/socket.sock\n2025-10-17 14:27:12.386 [info] Using configured platform linux for remote host horeka\n2025-10-17 14:27:12.391 [info] Using askpass script: /Users/mihir/.cursor/extensions/anysphere.remote-ssh-1.0.33/dist/scripts/launchSSHAskpass.sh with javascript file /Users/mihir/.cursor/extensions/anysphere.remote-ssh-1.0.33/dist/scripts/sshAskClient.js. Askpass handle: /var/folders/mr/xp86mpcd01sd72mft8vjq27r0000gn/T/cursor-ssh-Vf6muF/socket.sock\n2025-10-17 14:27:12.439 [info] Launching SSH server via shell with command: cat ""/var/folders/mr/xp86mpcd01sd72mft8vjq27r0000gn/T/cursor_remote_install_aaae7e95-0157-4d26-bbba-4e135dcf22c0.sh"" | ssh -T -D 59659 horeka bash --login -c bash\n2025-10-17 14:27:12.439 [info] Establishing SSH connection: cat ""/var/folders/mr/xp86mpcd01sd72mft8vjq27r0000gn/T/cursor_remote_install_aaae7e95-0157-4d26-bbba-4e135dcf22c0.sh"" | ssh -T -D 59659 horeka bash --login -c bash\n2025-10-17 14:27:12.440 [info] Started installation script. Waiting for it to finish...\n2025-10-17 14:27:12.440 [info] Waiting for server to install. Timeout: 30000ms\n2025-10-17 14:27:14.393 [info] Askpass server received request: POST /\n2025-10-17 14:27:14.395 [info] Askpass server received request body: {""request"":""([email protected]) Your OTP: ""}\n2025-10-17 14:27:14.395 [info] Pausing timeout; waiting for askpass response\n2025-10-17 14:27:14.395 [info] Received SSH askpass request: ([email protected]) Your OTP: \n2025-10-17 14:27:26.404 [info] Resuming timeout; askpass response received\n2025-10-17 14:27:27.710 [info] Askpass server received request: POST /\n2025-10-17 14:27:27.711 [info] Askpass server received request body: {""request"":""([email protected]) Password: ""}\n2025-10-17 14:27:27.711 [info] Pausing timeout; waiting for askpass response\n2025-10-17 14:27:27.711 [info] Received SSH askpass request: ([email protected]) Password: \n2025-10-17 14:27:30.791 [info] Resuming timeout; askpass response received\n2025-10-17 14:27:35.361 [info] (ssh_tunnel) stdout: Configuring Cursor Server on Remote\n\n2025-10-17 14:27:35.387 [info] (ssh_tunnel) stdout: Using TMP_DIR: /run/user/999226\n\n2025-10-17 14:27:35.580 [info] (ssh_tunnel) stdout: Locking /run/user/999226/cursor-remote-lock.41b851e3afada0dcdfba85e69b64011c\n\n2025-10-17 14:27:35.619 [info] (ssh_tunnel) stdout: Downloading server via wget from https://downloads.cursor.com/production/b9e5948c1ad20443a5cecba6b84a3c9b99d62582/linux/x64/cursor-reh-linux-x64.tar.gz to cursor-server-b1bbf4ae-e357-45d9-9f3e-0f589c7d9160.tar.gz\n\n2025-10-17 14:27:35.642 [info] (ssh_tunnel) stderr: --2025-10-17 14:27:35--  https://downloads.cursor.com/production/b9e5948c1ad20443a5cecba6b84a3c9b99d62582/linux/x64/cursor-reh-linux-x64.tar.gz\n\n2025-10-17 14:27:35.758 [info] (ssh_tunnel) stderr: Resolving downloads.cursor.com (downloads.cursor.com)... \n2025-10-17 14:27:35.782 [info] (ssh_tunnel) stderr: 2606:4700::6812:1080, 2606:4700::6812:1180, 104.18.16.128, ...\nConnecting to downloads.cursor.com (downloads.cursor.com)|2606:4700::6812:1080|:443... \n2025-10-17 14:27:35.783 [info] (ssh_tunnel) stderr: connected.\n\n2025-10-17 14:27:35.872 [info] (ssh_tunnel) stderr: HTTP request sent, awaiting response... \n2025-10-17 14:27:35.888 [info] (ssh_tunnel) stderr: 200 OK\nLength: 64733248 (62M) [application/gzip]\nSaving to: ‘cursor-server-b1bbf4ae-e357-45d9-9f3e-0f589c7d9160.tar.gz’\n\n\ncursor-server-b1bbf   0%[                    ]       0  --.-KB/s               \n2025-10-17 14:27:36.087 [info] (ssh_tunnel) stderr: \ncursor-server-b1bbf  79%[==============>     ]  49.19M   246MB/s               \n2025-10-17 14:27:36.111 [info] (ssh_tunnel) stderr: \ncursor-server-b1bbf 100%[===================>]  61.73M   275MB/s    in 0.2s    \n\n\n2025-10-17 14:27:36.111 [info] (ssh_tunnel) stderr: 2025-10-17 14:27:36 (275 MB/s) - ‘cursor-server-b1bbf4ae-e357-45d9-9f3e-0f589c7d9160.tar.gz’ saved [64733248/64733248]\n\n\n2025-10-17 14:27:36.113 [info] (ssh_tunnel) stdout: Extracting server contents from cursor-server-b1bbf4ae-e357-45d9-9f3e-0f589c7d9160.tar.gz\n\n2025-10-17 14:27:48.707 [info] (ssh_tunnel) stdout: Checking node executable\nv20.18.2\n\n2025-10-17 14:27:48.741 [info] (ssh_tunnel) stdout: Checking for running multiplex server: /home/hk-project-p0023960/tum_cte0515/.cursor-server/bin/multiplex-server/3ce73d09cffc8f33c6d911e972bd0f6dabbe3e26e810844be8060e6b10987db8.js\n\n2025-10-17 14:27:48.885 [info] (ssh_tunnel) stdout: Running multiplex server: \n\n2025-10-17 14:27:48.901 [info] (ssh_tunnel) stdout: Creating multiplex server token file /run/user/999226/cursor-remote-multiplex.token.41b851e3afada0dcdfba85e69b64011c.3ce73d09cffc8f33c6d911e972bd0f6dabbe3e26e810844be8060e6b10987db8\n\n2025-10-17 14:27:48.948 [info] (ssh_tunnel) stdout: Creating directory for multiplex server: /home/hk-project-p0023960/tum_cte0515/.cursor-server/bin/multiplex-server\nWriting multiplex server script to /home/hk-project-p0023960/tum_cte0515/.cursor-server/bin/multiplex-server/3ce73d09cffc8f33c6d911e972bd0f6dabbe3e26e810844be8060e6b10987db8.js\n\n2025-10-17 14:27:48.949 [info] (ssh_tunnel) stdout: Starting multiplex server: /home/hk-project-p0023960/tum_cte0515/.cursor-server/bin/b9e5948c1ad20443a5cecba6b84a3c9b99d62580/node /home/hk-project-p0023960/tum_cte0515/.cursor-server/bin/multiplex-server/3ce73d09cffc8f33c6d911e972bd0f6dabbe3e26e810844be8060e6b10987db8.js 1ee871c9-0446-4594-ae27-10cdda6cbdc8 0\n\n2025-10-17 14:27:48.959 [info] (ssh_tunnel) stdout: Multiplex server started with PID 3206117 and wrote pid to file /run/user/999226/cursor-remote-multiplex.pid.41b851e3afada0dcdfba85e69b64011c.3ce73d09cffc8f33c6d911e972bd0f6dabbe3e26e810844be8060e6b10987db8\n\n2025-10-17 14:27:48.959 [info] (ssh_tunnel) stdout: Reading multiplex server token file /run/user/999226/cursor-remote-multiplex.token.41b851e3afada0dcdfba85e69b64011c.3ce73d09cffc8f33c6d911e972bd0f6dabbe3e26e810844be8060e6b10987db8\n\n2025-10-17 14:27:48.961 [info] (ssh_tunnel) stdout: Multiplex server token file found\n\n2025-10-17 14:27:48.970 [info] (ssh_tunnel) stdout: Reading multiplex server log file /run/user/999226/cursor-remote-multiplex.log.41b851e3afada0dcdfba85e69b64011c.3ce73d09cffc8f33c6d911e972bd0f6dabbe3e26e810844be8060e6b10987db8\n\n2025-10-17 14:27:49.530 [info] (ssh_tunnel) stdout: Checking for code servers\n\n2025-10-17 14:27:49.688 [info] (ssh_tunnel) stdout: Code server script is not running\nCreating code server token file /run/user/999226/cursor-remote-code.token.41b851e3afada0dcdfba85e69b64011c\n\n2025-10-17 14:27:49.699 [info] (ssh_tunnel) stdout: Starting code server script /home/hk-project-p0023960/tum_cte0515/.cursor-server/bin/b9e5948c1ad20443a5cecba6b84a3c9b99d62580/bin/cursor-server --start-server --host=127.0.0.1 --port 0  --connection-token-file /run/user/999226/cursor-remote-code.token.41b851e3afada0dcdfba85e69b64011c --telemetry-level off --enable-remote-auto-shutdown --accept-server-license-terms &> /run/user/999226/cursor-remote-code.log.41b851e3afada0dcdfba85e69b64011c &\n\n2025-10-17 14:27:49.713 [info] (ssh_tunnel) stdout: Code server started with PID 3206229 and wrote pid to file /run/user/999226/cursor-remote-code.pid.41b851e3afada0dcdfba85e69b64011c\n\n2025-10-17 14:27:49.718 [info] (ssh_tunnel) stdout: Code server log file is /run/user/999226/cursor-remote-code.log.41b851e3afada0dcdfba85e69b64011c\n\n2025-10-17 14:27:50.269 [info] (ssh_tunnel) stdout: 2aef908d612b8a00fe727eb3: start\nexitCode==0==\nnodeExecutable==/home/hk-project-p0023960/tum_cte0515/.cursor-server/bin/b9e5948c1ad20443a5cecba6b84a3c9b99d62580/node==\nerrorMessage====\nisFatalError==false==\nmultiplexListeningOn==42623==\nmultiplexConnectionToken==1ee871c9-0446-4594-ae27-10cdda6cbdc8==\ncodeListeningOn==36077==\ncodeConnectionToken==ef447bc0-3351-48ed-9058-c609ca5ada42==\ndetectedPlatform==linux==\narch==x64==\nSSH_AUTH_SOCK====\n2aef908d612b8a00fe727eb3: end\n\n2025-10-17 14:27:50.282 [info] Server install command exit code:  0\n2025-10-17 14:27:50.282 [info] Deleting local script /var/folders/mr/xp86mpcd01sd72mft8vjq27r0000gn/T/cursor_remote_install_aaae7e95-0157-4d26-bbba-4e135dcf22c0.sh\n2025-10-17 14:27:50.285 [info] [forwarding][code] creating new forwarding server\n2025-10-17 14:27:50.285 [info] [forwarding][code] server listening on 127.0.0.1:59675\n2025-10-17 14:27:50.285 [info] [forwarding][code] Set up server\n2025-10-17 14:27:50.285 [info] [remote-ssh] codeListeningOn (remote=127.0.0.1:36077; local=127.0.0.1:59675) codeConnectionToken: ef447bc0-3351-48ed-9058-c609ca5ada42\n2025-10-17 14:27:50.286 [info] [forwarding][multiplex] creating new forwarding server\n2025-10-17 14:27:50.286 [info] [forwarding][multiplex] server listening on 127.0.0.1:59676\n2025-10-17 14:27:50.286 [info] [forwarding][multiplex] Set up server\n2025-10-17 14:27:50.291 [info] [remote-ssh] multiplexListeningOn (remote=[object Object]; local=[object Object]) multiplexConnectionToken: 1ee871c9-0446-4594-ae27-10cdda6cbdc8\n2025-10-17 14:27:50.291 [info] [remote-ssh] Pinging remote server via 127.0.0.1:59676...\n2025-10-17 14:27:50.305 [info] [remote-ssh] Resolved exec server. Socks port: 59659\n2025-10-17 14:27:50.306 [info] Setting up 0 default forwarded ports\n2025-10-17 14:27:50.306 [info] [remote-ssh] Resolved authority: {""host"":""127.0.0.1"",""port"":59675,""connectionToken"":""ef447bc0-3351-48ed-9058-c609ca5ada42"",""extensionHostEnv"":{}}. Socks port: 59659\n2025-10-17 14:27:50.312 [info] (ssh_tunnel) stdout: Unlocking /run/user/999226/cursor-remote-lock.41b851e3afada0dcdfba85e69b64011c\n \n***********************************************************************\n* This terminal is used to establish and maintain the SSH connection. *\n* Closing this terminal will terminate the connection and disconnect  *\n* Cursor from the remote server.                                      *\n***********************************************************************\n\n2025-10-17 14:27:50.321 [info] [forwarding][multiplex][127.0.0.1:59676 -> 127.0.0.1:42623][a3a6ebdb-cdda-4698-ba41-f0759f05c484] received connection request\n2025-10-17 14:27:50.323 [info] [command][161c1e8d-2172-4ae5-94c8-a366debebbe8] Sending command request: {""command"":""echo"",""args"":[""1""],""env"":{},""token"":""1ee871c9-0446-4594-ae27-10cdda6cbdc8"",""id"":""161c1e8d-2172-4ae5-94c8-a366debebbe8""}\n2025-10-17 14:27:50.348 [info] [forwarding][multiplex][127.0.0.1:59676 -> 127.0.0.1:59659 -> 127.0.0.1:42623][a3a6ebdb-cdda-4698-ba41-f0759f05c484] socks forwarding established\n2025-10-17 14:27:50.398 [info] [command][161c1e8d-2172-4ae5-94c8-a366debebbe8] Process exited with code 0\n2025-10-17 14:27:50.399 [info] [forwarding][multiplex][127.0.0.1:59676 -> 127.0.0.1:59659 -> 127.0.0.1:42623][a3a6ebdb-cdda-4698-ba41-f0759f05c484] socks connection closed\n2025-10-17 14:27:50.399 [info] [command][161c1e8d-2172-4ae5-94c8-a366debebbe8] Socket close event received\n2025-10-17 14:27:50.584 [info] [forwarding][code][127.0.0.1:59675 -> 127.0.0.1:36077][d1cd5e90-a9b8-4110-bf40-8ca1a14a1cd7] received connection request\n2025-10-17 14:27:50.608 [info] [forwarding][code][127.0.0.1:59675 -> 127.0.0.1:59659 -> 127.0.0.1:36077][d1cd5e90-a9b8-4110-bf40-8ca1a14a1cd7] socks forwarding established\n2025-10-17 14:27:50.650 [info] [forwarding][code][127.0.0.1:59675 -> 127.0.0.1:36077][23df6745-4ac6-4952-ae6d-8765dce6678e] received connection request\n2025-10-17 14:27:50.665 [info] [forwarding][code][127.0.0.1:59675 -> 127.0.0.1:59659 -> 127.0.0.1:36077][23df6745-4ac6-4952-ae6d-8765dce6678e] socks forwarding established\n2025-10-17 14:27:52.177 [info] Saved platform linux for remote host horeka\n",log,tab
+3,1993,"slurm/jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh",0,0,"",shellscript,tab
+4,2155,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"2:28:12 PM [info] Activating crowd-code\n2:28:12 PM [info] Recording started\n2:28:12 PM [info] Initializing git provider using file system watchers...\n2:28:13 PM [info] Git repository found\n2:28:13 PM [info] Git provider initialized successfully\n2:28:13 PM [info] Initial git state: [object Object]\n",Log,tab
+5,3541,"slurm/jobs/mihir/horeka/coinrun/speed_ablation/batch_size_2048/train_dyn_grain_ablation.sh",0,0,"",shellscript,tab
+6,3547,"TERMINAL",0,0,"",,terminal_focus
+7,17842,"/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/train_dynamics_maskgit_grain_ablation_3562055.log",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=05:00:00\n#SBATCH --partition=accelerated-h100\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:4\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/coinrun/dynamics/maskgit/%x_%j.log\n#SBATCH --job-name=train_dynamics_maskgit_grain_ablation\n#SBATCH --requeue\n#SBATCH --signal=b:usr1@300 # 5 min before timeout\n\n# --- signal trap to requeue job before timeout ---\nrequeue_job() {\n    echo ""[$(date)] caught sigusr1 (timeout warning), requeueing slurm job $SLURM_JOB_ID...""\n    # optional: trigger checkpoint saving here\n    # e.g., touch $checkpoint_dir/requeue_trigger\n    scontrol requeue $SLURM_JOB_ID\n    exit 0\n}\n\ntrap requeue_job sigusr1\n\n# set checkpoint flag based on restart count\nrestart_count=$(scontrol show job $SLURM_JOB_ID | grep -o 'Restarts=[0-9]*' | cut -d'=' -f2)\n\nif [ $restart_count -eq 0 ]; then\n    restore_ckpt_flag=""--no-restore-ckpt""\nelse\n    restore_ckpt_flag=""--restore-ckpt""\nfi\n\n\nexport CUDA_VISIBLE_DEVICES=0\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_coinrun/npy_test\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/coinrun/maskgit/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_ckpt_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/big-runs/tokenizer/train_tokenizer_default/3528955\n\nenv | grep SLURM\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    $restore_ckpt_flag \\n    --wandb_id $SLURM_JOB_ID \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --batch_size=1024 \\n    --patch_size=16 \\n    --warmup_steps=0 \\n    --wsd_decay_steps=0 \\n    --num_steps=10_000 \\n    --log_image_interval=100_000 \\n    --log \\n    --log_checkpoint_interval=100_000 \\n    --name=coinrun-dynamics-maskgit-grain-ablation-bs1024-$slurm_job_id \\n    --tags coinrun dynamics maskgit grain-ablation bs1024 \\n    --entity instant-uv \\n    --project jafar \\n    --tokenizer_checkpoint=$tokenizer_ckpt_dir \\n    --data_dir $array_records_dir_train &\n\nchild_pid=$!\n\nwait $child_pid/var/spool/slurmd/job3562055/slurm_script: line 43: .venv/bin/activate: No such file or directory\nSLURM_JOB_USER=tum_cte0515\nSLURM_TASKS_PER_NODE=1\nSLURM_JOB_UID=999226\nSLURM_TASK_PID=2451356\nSLURM_JOB_GPUS=0,1,2,3\nSLURM_LOCALID=0\nSLURM_SUBMIT_DIR=/hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine_jobs\nSLURMD_NODENAME=hkn0917\nSLURM_JOB_START_TIME=1760279968\nSLURM_CLUSTER_NAME=hk\nSLURM_JOB_END_TIME=1760297968\nSLURM_CPUS_ON_NODE=6\nSLURM_JOB_CPUS_PER_NODE=6\nSLURM_GPUS_ON_NODE=4\nSLURM_GTIDS=0\nSLURM_JOB_PARTITION=accelerated-h100\nSLURM_TRES_PER_TASK=cpu=5\nSLURM_OOM_KILL_STEP=0\nSLURM_JOB_NUM_NODES=1\nSLURM_JOBID=3562055\nSLURM_JOB_QOS=normal\nSLURM_PROCID=0\nSLURM_CPUS_PER_TASK=5\nSLURM_NTASKS=1\nSLURM_TOPOLOGY_ADDR=hkibb.hkibbi4.hkibbi4e2.hkn0917\nSLURM_TOPOLOGY_ADDR_PATTERN=switch.switch.switch.node\nSLURM_SCRIPT_CONTEXT=prolog_task\nSLURM_NODELIST=hkn0917\nSLURM_JOB_ACCOUNT=hk-project-p0023960\nSLURM_PRIO_PROCESS=0\nSLURM_NPROCS=1\nSLURM_NNODES=1\nSLURM_SUBMIT_HOST=hkn1993.localdomain\nSLURM_JOB_ID=3562055\nSLURM_NODEID=0\nSLURM_CONF=/etc/slurm/slurm.conf\nSLURM_JOB_NAME=train_dynamics_maskgit_grain_ablation\nSLURM_NTASKS_PER_NODE=1\nSLURM_JOB_GID=502226\nSLURM_JOB_NODELIST=hkn0917\nGpuFreq=control_disabled\nW1012 16:40:06.278108 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:06.620322 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:06.733623 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:06.806173 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:06.845624 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:06.920240 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:06.984605 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.020546 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.037437 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.076474 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.123455 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.268985 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.384836 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.424767 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.463488 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.501554 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.539940 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.556504 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.573998 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.612195 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.653245 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.693419 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.762899 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.800440 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.839500 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.877807 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.894711 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:07.913906 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.006532 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.181725 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.264634 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.303428 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.322088 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.344175 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.482400 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.602166 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.642301 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.788758 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:08.908572 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.107393 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.183351 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.222005 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.340740 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.408423 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.450425 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.494651 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.533221 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.675608 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.752636 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:09.980066 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.094695 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.293123 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.350819 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.387901 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.507092 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.575955 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.619004 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.662116 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:10.699876 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:11.053999 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:11.167886 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:11.207556 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:11.323145 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:11.425586 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:11.534228 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:11.591821 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nwandb: Currently logged in as: mihir-mahajan2002 (instant-uv) to https://api.wandb.ai. Use `wandb login --relogin` to force relogin\nwandb: creating run\nwandb: Tracking run with wandb version 0.22.0\nwandb: Run data is saved locally in /hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine_jobs/wandb/run-20251012_164011-3562055\nwandb: Run `wandb offline` to turn off syncing.\nwandb: Syncing run coinrun-dynamics-maskgit-grain-ablation-bs1024-3562055\nwandb: ⭐️ View project at https://wandb.ai/instant-uv/jafar\nwandb: 🚀 View run at https://wandb.ai/instant-uv/jafar/runs/3562055\nW1012 16:40:13.377784 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.416190 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.454586 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.493075 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.531568 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.569808 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.607048 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.644294 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.708505 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.746751 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.784558 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.840759 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.879445 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.916986 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:13.975334 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.033039 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.199132 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.216748 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.233481 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.250267 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.267058 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.283697 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.300518 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.317590 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.334499 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.351376 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.385624 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.403368 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.420082 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.436760 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.466665 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.483686 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.500812 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.517617 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.534747 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.551155 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.568010 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.598417 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.628794 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nW1012 16:40:14.763067 2451418 sol_gpu_cost_model.cc:102] No SoL config found for device: NVIDIA H100. Using default config.\nRunning on 1 devices.\nCounting all components: ['dynamics', 'lam', 'tokenizer']\nParameter counts:\n{'dynamics': 26555904, 'lam': 17640416, 'tokenizer': 34489696, 'total': 78686016}\nTraceback (most recent call last):\n  File ""/hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine_jobs/jasmine/train_dynamics.py"", line 821, in <module>\n    main(args)\n  File ""/hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine_jobs/jasmine/train_dynamics.py"", line 466, in main\n    train_iterator = build_dataloader(args, args.data_dir)\n                     ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^\n  File ""/hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine_jobs/jasmine/train_dynamics.py"", line 211, in build_dataloader\n    grain_dataloader = get_dataloader(\n                       ^^^^^^^^^^^^^^^\n  File ""/hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine_jobs/jasmine/utils/dataloader.py"", line 120, in get_dataloader\n    raise ValueError(""array_record_paths list cannot be empty."")\nValueError: array_record_paths list cannot be empty.\nwandb: \nwandb: 🚀 View run coinrun-dynamics-maskgit-grain-ablation-bs1024-3562055 at: https://wandb.ai/instant-uv/jafar/runs/3562055\nwandb: Find logs at: ../../../../../hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine_jobs/wandb/run-20251012_164011-3562055/logs\nW1012 16:40:15.850200 2451538 pjrt_client.cc:1469] WatchJobStateAsync failed for task goo.gle/debugproto    job_name: ""jax_worker"": CANCELLED: CANCELLED\nAdditional GRPC error information from remote target coordination_service while calling /tensorflow.CoordinationService/WatchJobState:\n:UNKNOWN:Error received from peer  {grpc_message:""CANCELLED"", grpc_status:1} [type.googleapis.com/tensorflow.DerivedStatus='']\nsrun: error: hkn0917: task 0: Exited with exit code 1\n\n============================= JOB FEEDBACK =============================\n\nJob ID: 3562055\nCluster: hk\nUser/Group: tum_cte0515/hk-project-p0023960\nAccount: hk-project-p0023960\nState: FAILED (exit code 1)\nPartition: accelerated-h100\nNodes: 1\nCores per node: 6\nNodelist: hkn0917\nCPU Utilized: 00:00:15\nCPU Efficiency: 5.10% of 00:04:54 core-walltime\nJob Wall-clock time: 00:00:49\nStarttime: Sun Oct 12 16:39:28 2025\nEndtime: Sun Oct 12 16:40:17 2025\nMemory Utilized: 3.60 GB\nMemory Efficiency: 0.48% of 750.00 GB (750.00 GB/node)\nEnergy Consumed: 20316 Joule / 5.64333333333333 Watthours\nAverage node power draw: 414.612244897959 Watt\n",log,tab
+8,320235,"slurm/jobs/mihir/horeka/doom/resolution60x80/train_dyn_default.sh",0,0,"#!/usr/bin/env bash\n\n#SBATCH --nodes=1\n#SBATCH --ntasks-per-node=1\n#SBATCH --time=48:00:00\n#SBATCH --partition=accelerated\n#SBATCH --cpus-per-task=5\n#SBATCH --gres=gpu:1\n#SBATCH --output=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/doom/dynamics/maskgit/%x_%j.log\n#SBATCH --error=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/logs/logs_mihir/doom/dynamics/maskgit/%x_%j.log\n#SBATCH --job-name=dynamics_doom_60x80\n#SBATCH --exclude=hkn0735\n#SBATCH --requeue\n#SBATCH --signal=b:usr1@300 # 5 min before timeout\n\n# --- signal trap to requeue job before timeout ---\nrequeue_job() {\n    echo ""[$(date)] caught sigusr1 (timeout warning), requeueing slurm job $SLURM_JOB_ID...""\n    # optional: trigger checkpoint saving here\n    # e.g., touch $checkpoint_dir/requeue_trigger\n    scontrol requeue $SLURM_JOB_ID\n    exit 0\n}\n\ntrap requeue_job sigusr1\n\n# set checkpoint flag based on restart count\nrestart_count=$(scontrol show job $SLURM_JOB_ID | grep -o 'Restarts=[0-9]*' | cut -d'=' -f2)\n\nif [ $restart_count -eq 0 ]; then\n    restore_ckpt_flag=""--no-restore-ckpt""\nelse\n    restore_ckpt_flag=""--restore-ckpt""\nfi\n\n\n\n# Log the sbatch script\ncat $0\n\nmodule unload mpi/openmpi/5.0\nmodule unload devel/cuda/12.4\nsource .venv/bin/activate\n\narray_records_dir_train=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom/doom_episodes_10m_60x80_fixed/train\narray_records_dir_val=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/data_doom/doom_episodes_10m_60x80_fixed/val\n\njob_name=$SLURM_JOB_NAME\nslurm_job_id=$SLURM_JOB_ID\n\nCHECKPOINT_DIR=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/doom/maskgit/$job_name/$slurm_job_id\nmkdir -p $CHECKPOINT_DIR\n\ntokenizer_ckpt_dir=/hkfs/work/workspace/scratch/tum_ind3695-jafa_ws_shared/checkpoints/mihir/tokenizer/train_tokenizer_default_single_gpu_60x80/3547697\n\nenv | grep SLURM\n\nsrun python jasmine/train_dynamics.py \\n    --save_ckpt \\n    $restore_ckpt_flag \\n    --wandb_id $SLURM_JOB_ID \\n    --num_actions=20 \\n    --ckpt_dir $CHECKPOINT_DIR \\n    --name=doom-dynamics-60x80-$slurm_job_id \\n    --image_height=60 \\n    --image_width=80 \\n    --tags doom dynamics maskgit default 60x80 \\n    --entity instant-uv \\n    --project jafar \\n    --tokenizer_checkpoint=$tokenizer_ckpt_dir \\n    --data_dir $array_records_dir_train \\n    --val_data_dir $array_records_dir_val &\n\nchild_pid=$!\n\nwait $child_pid",shellscript,tab
+9,1227671,"jasmine/train_tokenizer.py",0,0,"import os\n\nos.environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ""0.98"")\n\nfrom dataclasses import dataclass, field\nfrom typing import cast, Optional\n\nimport einops\nimport itertools\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax.checkpoint as ocp\nimport numpy as np\nimport dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\nimport grain\nimport flax.nnx as nnx\n\nfrom models.tokenizer import TokenizerVQVAE\nfrom utils.dataloader import get_dataloader\nfrom utils.train_utils import (\n    get_lr_schedule,\n    count_parameters_by_component,\n    print_mem_stats,\n    print_compiled_memory_stats,\n    print_compiled_cost_analysis,\n)\n\n\n@dataclass\nclass Args:\n    # Experiment\n    num_steps: int = 300_000\n    seed: int = 0\n    seq_len: int = 16\n    image_channels: int = 3\n    image_height: int = 64\n    image_width: int = 64\n    data_dir: str = """"\n    save_ckpt: bool = False\n    restore_ckpt: bool = False\n    # Optimization\n    vq_beta: float = 0.25\n    batch_size: int = 48\n    init_lr: float = 0.0\n    max_lr: float = 3e-4\n    decay_end: float = 0.0\n    wsd_decay_steps: int = (\n        30_000  # NOTE: wsd_decay_steps will only be used when using a wsd-schedule\n    )\n    lr_schedule: str = ""wsd""  # supported options: wsd, cos\n    warmup_steps: int = 10000\n    # Tokenizer\n    model_dim: int = 512\n    ffn_dim: int = 2048\n    latent_dim: int = 32\n    num_latents: int = 1024\n    patch_size: int = 16\n    num_blocks: int = 4\n    num_heads: int = 8\n    dropout: float = 0.0\n    codebook_dropout: float = 0.01\n    param_dtype = jnp.float32\n    dtype = jnp.bfloat16\n    use_flash_attention: bool = True\n    # Logging\n    log: bool = True\n    entity: str = """"\n    project: str = """"\n    name: str = ""train_tokenizer""\n    tags: list[str] = field(default_factory=lambda: [""tokenizer""])\n    log_interval: int = 50\n    log_image_interval: int = 1000\n    ckpt_dir: str = """"\n    log_checkpoint_interval: int = 1000\n    log_checkpoint_keep_period: int = 20_000\n    log_gradients: bool = False\n    val_data_dir: str = """"\n    val_interval: int = 20_000\n    val_steps: int = 50\n    wandb_id: str = """"\n\n\ndef build_model(args: Args, rng: jax.Array) -> tuple[TokenizerVQVAE, jax.Array]:\n    rng, _rng = jax.random.split(rng)\n    rngs = nnx.Rngs(_rng)\n    return (\n        TokenizerVQVAE(\n            in_dim=args.image_channels,\n            model_dim=args.model_dim,\n            ffn_dim=args.ffn_dim,\n            latent_dim=args.latent_dim,\n            num_latents=args.num_latents,\n            patch_size=args.patch_size,\n            num_blocks=args.num_blocks,\n            num_heads=args.num_heads,\n            dropout=args.dropout,\n            codebook_dropout=args.codebook_dropout,\n            param_dtype=args.param_dtype,\n            dtype=args.dtype,\n            use_flash_attention=args.use_flash_attention,\n            rngs=rngs,\n        ),\n        rng,\n    )\n\n\ndef build_optimizer(model: TokenizerVQVAE, args: Args) -> nnx.ModelAndOptimizer:\n    lr_schedule = get_lr_schedule(\n        args.lr_schedule,\n        args.init_lr,\n        args.max_lr,\n        args.decay_end,\n        args.num_steps,\n        args.warmup_steps,\n        args.wsd_decay_steps,\n    )\n    tx = optax.adamw(\n        learning_rate=lr_schedule,\n        b1=0.9,\n        b2=0.9,\n        weight_decay=1e-4,\n        mu_dtype=args.param_dtype,  # moments in full precision\n    )\n    optimizer = nnx.ModelAndOptimizer(model, tx)\n    return optimizer\n\n\ndef build_mesh_and_sharding(\n    num_devices: int,\n) -> tuple[Mesh, NamedSharding, NamedSharding]:\n    device_mesh_arr = create_device_mesh((num_devices,))\n    mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n    replicated_sharding = NamedSharding(mesh, PartitionSpec())\n    videos_sharding = NamedSharding(mesh, PartitionSpec(""data"", None, None, None, None))\n    return mesh, replicated_sharding, videos_sharding\n\n\ndef shard_optimizer_states(\n    optimizer: nnx.ModelAndOptimizer, replicated_sharding: NamedSharding\n) -> None:\n    model_state = nnx.state(optimizer.model)\n    model_sharded_state = jax.lax.with_sharding_constraint(\n        model_state, replicated_sharding\n    )\n    nnx.update(optimizer.model, model_sharded_state)\n    optimizer_state = nnx.state(optimizer, nnx.optimizer.OptState)\n    optimizer_sharded_state = jax.lax.with_sharding_constraint(\n        optimizer_state, replicated_sharding\n    )\n    nnx.update(optimizer, optimizer_sharded_state)\n\n\ndef build_dataloader(args: Args, data_dir: str) -> grain.DataLoaderIterator:\n    image_shape = (args.image_height, args.image_width, args.image_channels)\n    array_record_files = [\n        os.path.join(data_dir, x)\n        for x in os.listdir(data_dir)\n        if x.endswith("".array_record"")\n    ]\n    grain_dataloader = get_dataloader(\n        array_record_files,\n        args.seq_len,\n        # NOTE: We deliberately pass the global batch size\n        # The dataloader shards the dataset across all processes\n        args.batch_size,\n        *image_shape,\n        num_workers=8,\n        prefetch_buffer_size=1,\n        seed=args.seed,\n    )\n    initial_state = grain_dataloader._create_initial_state()\n    grain_iterator = grain.DataLoaderIterator(grain_dataloader, initial_state)\n    return grain_iterator\n\n\ndef build_checkpoint_manager(args: Args) -> Optional[ocp.CheckpointManager]:\n    if args.restore_ckpt or args.save_ckpt:\n        handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n        handler_registry.add(\n            ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n        )\n        handler_registry.add(\n            ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n        )\n        handler_registry.add(\n            ""train_dataloader_state"",\n            grain.checkpoint.CheckpointSave,\n            cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n        )\n        handler_registry.add(\n            ""train_dataloader_state"",\n            grain.checkpoint.CheckpointRestore,\n            cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n        )\n        if args.val_data_dir:\n            handler_registry.add(\n                ""val_dataloader_state"",\n                grain.checkpoint.CheckpointSave,\n                cast(\n                    ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler\n                ),\n            )\n            handler_registry.add(\n                ""val_dataloader_state"",\n                grain.checkpoint.CheckpointRestore,\n                cast(\n                    ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler\n                ),\n            )\n        checkpoint_options = ocp.CheckpointManagerOptions(\n            save_interval_steps=args.log_checkpoint_interval,\n            max_to_keep=3,\n            keep_period=args.log_checkpoint_keep_period,\n            step_format_fixed_length=6,\n            cleanup_tmp_directories=True,\n        )\n        checkpoint_manager = ocp.CheckpointManager(\n            args.ckpt_dir,\n            options=checkpoint_options,\n            handler_registry=handler_registry,\n        )\n        return checkpoint_manager\n    else:\n        return None\n\n\ndef restore_checkpoint_if_needed(\n    args: Args,\n    checkpoint_manager: Optional[ocp.CheckpointManager],\n    optimizer: nnx.ModelAndOptimizer,\n    train_iterator: grain.DataLoaderIterator,\n    val_iterator: Optional[grain.DataLoaderIterator],\n    restore_step: Optional[int] = None,\n) -> tuple[\n    int, nnx.ModelAndOptimizer, grain.DataLoaderIterator, grain.DataLoaderIterator\n]:\n    step = 0\n    if checkpoint_manager and restore_step is None:\n        restore_step = checkpoint_manager.latest_step()\n    if args.restore_ckpt:\n        assert checkpoint_manager is not None\n        abstract_optimizer = nnx.eval_shape(lambda: optimizer)\n        abstract_optimizer_state = nnx.state(abstract_optimizer)\n        if val_iterator:\n            restore_args = ocp.args.Composite(\n                model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n                train_dataloader_state=grain.checkpoint.CheckpointRestore(train_iterator),  # type: ignore\n                val_dataloader_state=grain.checkpoint.CheckpointRestore(val_iterator),  # type: ignore\n            )\n        else:\n            restore_args = ocp.args.Composite(\n                model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n                train_dataloader_state=grain.checkpoint.CheckpointRestore(train_iterator),  # type: ignore\n            )\n        restored = checkpoint_manager.restore(restore_step, args=restore_args)\n        restored_optimizer_state = restored[""model_state""]\n        nnx.update(optimizer, restored_optimizer_state)\n        train_iterator = restored[""train_dataloader_state""]\n        if val_iterator:\n            val_iterator = restored[""val_dataloader_state""]\n        step = restore_step or 0\n        print(f""Restored dataloader and model state from step {step}"")\n    return step, optimizer, train_iterator, val_iterator\n\n\ndef main(args: Args) -> None:\n    jax.distributed.initialize()\n    num_devices = jax.device_count()\n    if num_devices == 0:\n        raise ValueError(""No JAX devices found."")\n    print(f""Running on {num_devices} devices."")\n\n    if args.batch_size % num_devices != 0:\n        raise ValueError(\n            f""Global batch size {args.batch_size} must be divisible by ""\n            f""number of devices {num_devices}.""\n        )\n\n    rng = jax.random.key(args.seed)\n\n    # --- Initialize model ---\n    tokenizer, rng = build_model(args, rng)\n\n    _, params, _ = nnx.split(tokenizer, nnx.Param, ...)\n    param_counts = count_parameters_by_component(params)\n\n    if args.log and jax.process_index() == 0:\n        wandb_init_kwargs = {\n            ""entity"": args.entity,\n            ""project"": args.project,\n            ""name"": args.name,\n            ""tags"": args.tags,\n            ""group"": ""debug"",\n            ""config"": args,\n        }\n\n        if args.wandb_id:\n            wandb_init_kwargs.update(\n                {\n                    ""id"": args.wandb_id,\n                    ""resume"": ""allow"",\n                }\n            )\n        wandb.init(**wandb_init_kwargs)\n\n        wandb.config.update({""model_param_count"": param_counts})\n\n    print(""Parameter counts:"")\n    print(param_counts)\n\n    # --- Initialize optimizer ---\n    optimizer = build_optimizer(tokenizer, args)\n    del tokenizer\n\n    # FIXME: switch to create_hybrid_device_mesh for runs spanning multiple nodes\n    _, replicated_sharding, videos_sharding = build_mesh_and_sharding(num_devices)\n\n    shard_optimizer_states(optimizer, replicated_sharding)\n\n    # --- Initialize checkpoint manager ---\n    checkpoint_manager = build_checkpoint_manager(args)\n\n    # --- Create DataLoaderIterator from dataloader ---\n    train_iterator = build_dataloader(args, args.data_dir)\n    val_iterator = None\n    if args.val_data_dir:\n        val_iterator = build_dataloader(args, args.val_data_dir)\n\n    # --- Restore checkpoint ---\n    step, optimizer, train_iterator, val_iterator = restore_checkpoint_if_needed(\n        args, checkpoint_manager, optimizer, train_iterator, val_iterator\n    )\n\n    # --- Define loss and train step (close over args) ---\n    def tokenizer_loss_fn(\n        model: TokenizerVQVAE, inputs: dict, training: bool = False\n    ) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n        gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n        inputs[""videos""] = gt.astype(args.dtype)\n        outputs = model(inputs, training=training)\n        outputs[""recon""] = outputs[""recon""].astype(jnp.float32)\n        mse = jnp.square(gt - outputs[""recon""]).mean()\n        q_loss = jnp.square(jax.lax.stop_gradient(outputs[""emb""]) - outputs[""z""]).mean()\n        commitment_loss = jnp.square(\n            outputs[""emb""] - jax.lax.stop_gradient(outputs[""z""])\n        ).mean()\n        loss = mse + q_loss + args.vq_beta * commitment_loss\n\n        gt_clipped = gt.clip(0, 1).reshape(-1, *gt.shape[2:])\n        recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n        psnr = jnp.asarray(pix.psnr(gt_clipped, recon)).mean()\n        ssim = jnp.asarray(pix.ssim(gt_clipped, recon)).mean()\n        _, index_counts = jnp.unique_counts(\n            jnp.ravel(outputs[""indices""]), size=args.num_latents, fill_value=0\n        )\n        codebook_usage = (index_counts != 0).mean()\n        metrics = dict(\n            loss=loss,\n            mse=mse,\n            q_loss=q_loss,\n            commitment_loss=commitment_loss,\n            psnr=psnr,\n            ssim=ssim,\n            codebook_usage=codebook_usage,\n        )\n        return loss, (outputs[""recon""], metrics)\n\n    @nnx.jit(donate_argnums=0)\n    def train_step(\n        optimizer: nnx.ModelAndOptimizer, inputs: dict\n    ) -> tuple[jax.Array, jax.Array, dict]:\n        def loss_fn(model: TokenizerVQVAE) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n            model.train()\n            return tokenizer_loss_fn(model, inputs, training=True)\n\n        (loss, (recon, metrics)), grads = nnx.value_and_grad(loss_fn, has_aux=True)(\n            optimizer.model\n        )\n        optimizer.update(grads)\n        if args.log_gradients:\n            metrics[""encoder_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""encoder""]\n            )\n            metrics[""vq_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""vq""]\n            )\n            metrics[""decoder_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""decoder""]\n            )\n        return loss, recon, metrics\n\n    @nnx.jit\n    def val_step(\n        tokenizer: TokenizerVQVAE, inputs: dict\n    ) -> tuple[jax.Array, jax.Array, dict]:\n        tokenizer.eval()\n        (loss, (recon, metrics)) = tokenizer_loss_fn(tokenizer, inputs, training=False)\n        return loss, recon, metrics\n\n    def calculate_validation_metrics(val_dataloader, tokenizer):\n        step = 0\n        loss_per_step = []\n        metrics_per_step = []\n        batch = None\n        recon = None\n        for batch in val_dataloader:\n            loss, recon, metrics = val_step(tokenizer, batch)\n            loss_per_step.append(loss)\n            metrics_per_step.append(metrics)\n            step += 1\n            if step > args.val_steps:\n                break\n\n        if step < args.val_steps:\n            print(\n                f""Warning: Your validation dataset is too small to make val_steps many steps. Made {step} steps, expected {args.val_steps}""\n            )\n\n        val_loss = np.mean(loss_per_step)\n        val_metrics = {\n            f""val_{key}"": np.mean([float(m[key]) for m in metrics_per_step])\n            for key in metrics_per_step[0].keys()\n        }\n        val_metrics[""val_loss""] = val_loss\n        return val_metrics, batch, recon\n\n    # --- TRAIN LOOP ---\n    dataloader_train = (\n        {\n            ""videos"": jax.make_array_from_process_local_data(\n                videos_sharding, elem[""videos""]\n            ),\n        }\n        for elem in train_iterator\n    )\n    dataloader_val = None\n    if val_iterator:\n        dataloader_val = (\n            {\n                ""videos"": jax.make_array_from_process_local_data(\n                    videos_sharding, elem[""videos""]\n                ),\n            }\n            for elem in val_iterator\n        )\n    if jax.process_index() == 0:\n        first_batch = next(dataloader_train)\n        compiled = train_step.lower(optimizer, first_batch).compile()\n        print_compiled_memory_stats(compiled.memory_analysis())\n        print_compiled_cost_analysis(compiled.cost_analysis())\n        # Do not skip the first batch during training\n        dataloader_train = itertools.chain([first_batch], dataloader_train)\n    print(f""Starting training from step {step}..."")\n    first_step = step\n    while step < args.num_steps:\n        for batch in dataloader_train:\n            # --- Train step ---\n            loss, recon, metrics = train_step(optimizer, batch)\n            if step == first_step:\n                print_mem_stats(""After params initialized"")\n            step += 1\n\n            # --- Validation loss ---\n            val_results = {}\n            if dataloader_val and step % args.val_interval == 0:\n                print(""Calculating validation metrics..."")\n                val_metrics, val_gt_batch, val_recon = calculate_validation_metrics(\n                    dataloader_val, optimizer.model\n                )\n                print(f""Step {step}, validation loss: {val_metrics['val_loss']}"")\n                val_results = {\n                    ""metrics"": val_metrics,\n                    ""gt_batch"": val_gt_batch,\n                    ""recon"": val_recon,\n                }\n\n            # --- Logging ---\n            if args.log:\n                if step % args.log_interval == 0 and jax.process_index() == 0:\n                    log_dict = {""loss"": loss, ""step"": step, **metrics}\n                    if val_results:\n                        log_dict.update(val_results[""metrics""])\n                    wandb.log(log_dict)\n                if step % args.log_image_interval == 0:\n                    gt_seq = batch[""videos""][0].astype(jnp.float32) / 255.0\n                    recon_seq = recon[0].clip(0, 1)\n                    comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n                    comparison_seq = einops.rearrange(\n                        comparison_seq * 255, ""t h w c -> h (t w) c""\n                    )\n                    if val_results and step % args.val_interval == 0:\n                        val_results[""gt_seq_val""] = (\n                            val_results[""gt_batch""][""videos""][0].astype(jnp.float32)\n                            / 255.0\n                        )\n                        val_results[""recon_seq_val""] = val_results[""recon""][0].clip(\n                            0, 1\n                        )\n                        val_results[""val_comparison_seq""] = jnp.concatenate(\n                            (val_results[""gt_seq_val""], val_results[""recon_seq_val""]),\n                            axis=1,\n                        )\n                        val_results[""val_comparison_seq""] = einops.rearrange(\n                            val_results[""val_comparison_seq""] * 255,\n                            ""t h w c -> h (t w) c"",\n                        )\n                    # NOTE: Process-dependent control flow deliberately happens\n                    # after indexing operation since it must not contain code\n                    # sections that lead to cross-accelerator communication.\n                    if jax.process_index() == 0:\n                        log_images = dict(\n                            image=wandb.Image(np.asarray(gt_seq[0])),\n                            recon=wandb.Image(np.asarray(recon_seq[0])),\n                            true_vs_recon=wandb.Image(\n                                np.asarray(comparison_seq.astype(np.uint8))\n                            ),\n                        )\n                        if val_results and step % args.val_interval == 0:\n                            log_images.update(\n                                dict(\n                                    val_image=wandb.Image(\n                                        np.asarray(val_results[""gt_seq_val""][0])\n                                    ),\n                                    val_recon=wandb.Image(\n                                        np.asarray(val_results[""recon_seq_val""][0])\n                                    ),\n                                    val_true_vs_recon=wandb.Image(\n                                        np.asarray(\n                                            val_results[""val_comparison_seq""].astype(\n                                                np.uint8\n                                            )\n                                        )\n                                    ),\n                                )\n                            )\n                        wandb.log(log_images)\n            # --- Checkpointing ---\n            if args.save_ckpt and step % args.log_checkpoint_interval == 0:\n                assert checkpoint_manager is not None\n                optimizer_state = nnx.state(optimizer)\n                if val_iterator:\n                    ckpt_manager_args = ocp.args.Composite(\n                        model_state=ocp.args.PyTreeSave(optimizer_state),  # type: ignore\n                        train_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            train_iterator  # type: ignore\n                        ),\n                        val_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            val_iterator  # type: ignore\n                        ),\n                    )\n                else:\n                    ckpt_manager_args = ocp.args.Composite(\n                        model_state=ocp.args.PyTreeSave(optimizer_state),  # type: ignore\n                        train_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            train_iterator  # type: ignore\n                        ),\n                    )\n                checkpoint_manager.save(step, args=ckpt_manager_args)\n                print(f""Saved checkpoint at step {step}"")\n            if step >= args.num_steps:\n                break\n\n    if checkpoint_manager:\n        checkpoint_manager.close()\n\n\nif __name__ == ""__main__"":\n    args = tyro.cli(Args)\n    main(args)\n",python,tab
+10,1229571,"jasmine/train_tokenizer.py",10,0,"",python,selection_mouse
+11,1230445,"jasmine/train_tokenizer.py",16,0,"",python,selection_mouse
+12,1230696,"jasmine/train_tokenizer.py",14,7,"environ",python,selection_mouse
+13,1230848,"jasmine/train_tokenizer.py",14,18,"environ.setdefault",python,selection_mouse
+14,1230848,"jasmine/train_tokenizer.py",14,50,"environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION",python,selection_mouse
+15,1230877,"jasmine/train_tokenizer.py",14,53,"environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ",python,selection_mouse
+16,1230878,"jasmine/train_tokenizer.py",14,58,"environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ""0.98",python,selection_mouse
+17,1230902,"jasmine/train_tokenizer.py",14,59,"environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ""0.98""",python,selection_mouse
+18,1230930,"jasmine/train_tokenizer.py",14,60,"environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ""0.98"")",python,selection_mouse
+19,1231261,"jasmine/train_tokenizer.py",74,0,"",python,selection_mouse
+20,1231267,"jasmine/train_tokenizer.py",73,0,"",python,selection_command
+21,1232099,"jasmine/train_tokenizer.py",68,0,"",python,selection_mouse
+22,1232416,"jasmine/train_tokenizer.py",72,0,"",python,selection_mouse
+23,1232849,"jasmine/train_tokenizer.py",10,0,"",python,selection_mouse

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-ea3d4740-7761-4d5b-b6ed-89bdaaf7e91f1760906482985-2025_10_19-22.42.16.510/source.csv

@@ -0,0 +1,69 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+2,2933,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"10:42:16 PM [info] Activating crowd-code\n10:42:16 PM [info] Recording started\n10:42:16 PM [info] Initializing git provider using file system watchers...\n10:42:17 PM [info] Git repository found\n10:42:17 PM [info] Git provider initialized successfully\n10:42:18 PM [info] Initial git state: [object Object]\n",Log,tab
+3,5491,"TERMINAL",0,0,"bash",,terminal_focus
+4,8092,"TERMINAL",0,0,"uv pip show",,terminal_command
+5,8147,"TERMINAL",0,0,"]633;C",,terminal_output
+6,8195,"TERMINAL",0,0,"warning: Please provide a package name or names.\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine",,terminal_output
+7,13077,"TERMINAL",0,0,"cd data/",,terminal_command
+8,13479,"TERMINAL",0,0,"ls",,terminal_command
+9,13488,"TERMINAL",0,0,"]633;Cjasmine_data  pyproject.toml  uv.lock  _vizdoom  _vizdoom.ini\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+10,17201,"TERMINAL",0,0,"source .venv/bin/activate",,terminal_command
+11,17253,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+12,21685,"TERMINAL",0,0,"uv pip freeze",,terminal_command
+13,22010,"TERMINAL",0,0,"]633;Cabsl-py==2.1.0\r\naiohappyeyeballs==2.4.0\r\naiohttp==3.10.5\r\naiosignal==1.3.1\r\nale-py==0.9.0\r\nappnope==0.1.4\r\narray-record==0.8.1\r\nasttokens==2.4.1\r\nasync-timeout==4.0.3\r\nattrs==24.2.0\r\nautorom==0.4.2\r\nautorom-accept-rom-license==0.6.1\r\nblack==24.8.0\r\nbox2d-py==2.3.5\r\ncertifi==2024.8.30\r\ncffi==1.17.1\r\ncharset-normalizer==3.3.2\r\nclick==8.1.7\r\ncloudpickle==3.0.0\r\ncomm==0.2.2\r\ncontourpy==1.3.0\r\ncramjam==2.8.3\r\ncycler==0.12.1\r\ncython==3.0.11\r\ndatasets==3.0.0\r\ndebugpy==1.8.5\r\ndecorator==4.4.2\r\ndill==0.3.8\r\ndocstring-parser==0.17.0\r\netils==1.13.0\r\nexceptiongroup==1.2.2\r\nexecuting==2.1.0\r\nfarama-notifications==0.0.4\r\nfasteners==0.19\r\nfastparquet==2024.5.0\r\nffmpeg-python==0.2.0\r\nfilelock==3.16.0\r\nfonttools==4.53.1\r\nfrozenlist==1.4.1\r\nfsspec==2024.6.1\r\nfuture==1.0.0\r\nglcontext==3.0.0\r\nglfw==2.7.0\r\ngrpcio==1.66.1\r\ngym==0.26.2\r\ngym-notices==0.0.8\r\ngym3==0.3.3\r\ngymnasium==0.29.1\r\nhf-transfer==0.1.9\r\nhf-xet==1.1.10\r\nhuggingface-hub==0.24.7\r\nidna==3.8\r\nimageio==2.35.1\r\nimageio-ffmpeg==0.5.1\r\nimportlib-metadata==8.5.0\r\nimportlib-resources==6.4.5\r\niniconfig==2.0.0\r\ninquirerpy==0.3.4\r\nipykernel==6.29.5\r\nipython==8.18.1\r\nipywidgets==8.1.5\r\n-e file:///hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine/data\r\njedi==0.19.1\r\njinja2==3.1.4\r\njupyter-client==8.6.2\r\njupyter-core==5.7.2\r\njupyterlab-widgets==3.0.13\r\nkiwisolver==1.4.7\r\nloguru==0.7.3\r\nlz4==4.3.3\r\nmarkdown==3.7\r\nmarkdown-it-py==3.0.0\r\nmarkupsafe==2.1.5\r\nmatplotlib==3.9.2\r\nmatplotlib-inline==0.1.7\r\nmdurl==0.1.2\r\nmoderngl==5.12.0\r\nmoviepy==1.0.3\r\nmpmath==1.3.0\r\nmujoco==3.2.5\r\nmujoco-py==2.1.2.14\r\nmultidict==6.1.0\r\nmultiprocess==0.70.16\r\nmypy-extensions==1.0.0\r\nnest-asyncio==1.6.0\r\nnetworkx==3.2.1\r\nnumpy==2.0.2\r\nnvidia-cublas-cu12==12.1.3.1\r\nnvidia-cuda-cupti-cu12==12.1.105\r\nnvidia-cuda-nvrtc-cu12==12.1.105\r\nnvidia-cuda-runtime-cu12==12.1.105\r\nnvidia-cudnn-cu12==9.1.0.70\r\nnvidia-cufft-cu12==11.0.2.54\r\nnvidia-curand-cu12==10.3.2.106\r\nnvidia-cusolver-cu12==11.4.5.107\r\nnvidia-cusparse-cu12==12.1.0.106\r\nnvidia-nccl-cu12==2.20.5\r\nnvidia-nvjitlink-cu12==12.9.86\r\nnvidia-nvtx-cu12==12.1.105\r\nopencv-python==4.10.0.84\r\npackaging==24.1\r\npandas==2.2.2\r\nparso==0.8.4\r\npathspec==0.12.1\r\npexpect==4.9.0\r\npfzy==0.3.4\r\npillow==10.4.0\r\nplatformdirs==4.3.2\r\npluggy==1.5.0\r\nprocgen==0.10.7\r\nproglog==0.1.10\r\nprompt-toolkit==3.0.47\r\nprotobuf==5.28.1\r\npsutil==6.0.0\r\nptyprocess==0.7.0\r\npure-eval==0.2.3\r\npy==1.11.0\r\npyarrow==17.0.0\r\npycparser==2.22\r\npygame==2.6.1\r\npygments==2.18.0\r\npyopengl==3.1.7\r\npyparsing==3.1.4\r\npytest==7.0.1\r\npython-dateutil==2.9.0.post0\r\npytz==2024.2\r\npyyaml==6.0.2\r\npyzmq==26.2.0\r\nrequests==2.32.3\r\nrich==13.8.1\r\nsetuptools==80.9.0\r\nshimmy==1.3.0\r\nshtab==1.7.2\r\nsix==1.16.0\r\nstable-baselines3==2.3.2\r\nstack-data==0.6.3\r\nswig==4.2.1\r\nsympy==1.13.2\r\ntensorboard==2.17.1\r\ntensorboard-data-server==0.7.2\r\ntomli==2.0.1\r\ntorch==2.4.1\r\ntornado==6.4.1\r\ntqdm==4.66.5\r\ntraitlets==5.14.3\r\ntriton==3.0.0\r\ntypeguard==4.4.4\r\ntyping-extensions==4.12.2\r\ntyro==0.9.32\r\ntzdata==2024.1\r\nurllib3==2.2.2\r\nvizdoom==1.2.4\r\nwcwidth==0.2.13\r\nwerkzeug==3.0.4\r\nwidgetsnbextension==4.0.13\r\nxxhash==3.5.0\r\nyarl==1.11.1\r\nzipp==3.20.1\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+14,22019,"TERMINAL",0,0,"",,terminal_output
+15,44759,"TERMINAL",0,0,"uv pip freeze > requirements_doom.txt",,terminal_command
+16,44764,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+17,61615,"data/requirements_doom.txt",0,0,"absl-py==2.1.0\naiohappyeyeballs==2.4.0\naiohttp==3.10.5\naiosignal==1.3.1\nale-py==0.9.0\nappnope==0.1.4\narray-record==0.8.1\nasttokens==2.4.1\nasync-timeout==4.0.3\nattrs==24.2.0\nautorom==0.4.2\nautorom-accept-rom-license==0.6.1\nblack==24.8.0\nbox2d-py==2.3.5\ncertifi==2024.8.30\ncffi==1.17.1\ncharset-normalizer==3.3.2\nclick==8.1.7\ncloudpickle==3.0.0\ncomm==0.2.2\ncontourpy==1.3.0\ncramjam==2.8.3\ncycler==0.12.1\ncython==3.0.11\ndatasets==3.0.0\ndebugpy==1.8.5\ndecorator==4.4.2\ndill==0.3.8\ndocstring-parser==0.17.0\netils==1.13.0\nexceptiongroup==1.2.2\nexecuting==2.1.0\nfarama-notifications==0.0.4\nfasteners==0.19\nfastparquet==2024.5.0\nffmpeg-python==0.2.0\nfilelock==3.16.0\nfonttools==4.53.1\nfrozenlist==1.4.1\nfsspec==2024.6.1\nfuture==1.0.0\nglcontext==3.0.0\nglfw==2.7.0\ngrpcio==1.66.1\ngym==0.26.2\ngym-notices==0.0.8\ngym3==0.3.3\ngymnasium==0.29.1\nhf-transfer==0.1.9\nhf-xet==1.1.10\nhuggingface-hub==0.24.7\nidna==3.8\nimageio==2.35.1\nimageio-ffmpeg==0.5.1\nimportlib-metadata==8.5.0\nimportlib-resources==6.4.5\niniconfig==2.0.0\ninquirerpy==0.3.4\nipykernel==6.29.5\nipython==8.18.1\nipywidgets==8.1.5\n-e file:///hkfs/home/project/hk-project-p0023960/tum_cte0515/Projects/jasmine/data\njedi==0.19.1\njinja2==3.1.4\njupyter-client==8.6.2\njupyter-core==5.7.2\njupyterlab-widgets==3.0.13\nkiwisolver==1.4.7\nloguru==0.7.3\nlz4==4.3.3\nmarkdown==3.7\nmarkdown-it-py==3.0.0\nmarkupsafe==2.1.5\nmatplotlib==3.9.2\nmatplotlib-inline==0.1.7\nmdurl==0.1.2\nmoderngl==5.12.0\nmoviepy==1.0.3\nmpmath==1.3.0\nmujoco==3.2.5\nmujoco-py==2.1.2.14\nmultidict==6.1.0\nmultiprocess==0.70.16\nmypy-extensions==1.0.0\nnest-asyncio==1.6.0\nnetworkx==3.2.1\nnumpy==2.0.2\nnvidia-cublas-cu12==12.1.3.1\nnvidia-cuda-cupti-cu12==12.1.105\nnvidia-cuda-nvrtc-cu12==12.1.105\nnvidia-cuda-runtime-cu12==12.1.105\nnvidia-cudnn-cu12==9.1.0.70\nnvidia-cufft-cu12==11.0.2.54\nnvidia-curand-cu12==10.3.2.106\nnvidia-cusolver-cu12==11.4.5.107\nnvidia-cusparse-cu12==12.1.0.106\nnvidia-nccl-cu12==2.20.5\nnvidia-nvjitlink-cu12==12.9.86\nnvidia-nvtx-cu12==12.1.105\nopencv-python==4.10.0.84\npackaging==24.1\npandas==2.2.2\nparso==0.8.4\npathspec==0.12.1\npexpect==4.9.0\npfzy==0.3.4\npillow==10.4.0\nplatformdirs==4.3.2\npluggy==1.5.0\nprocgen==0.10.7\nproglog==0.1.10\nprompt-toolkit==3.0.47\nprotobuf==5.28.1\npsutil==6.0.0\nptyprocess==0.7.0\npure-eval==0.2.3\npy==1.11.0\npyarrow==17.0.0\npycparser==2.22\npygame==2.6.1\npygments==2.18.0\npyopengl==3.1.7\npyparsing==3.1.4\npytest==7.0.1\npython-dateutil==2.9.0.post0\npytz==2024.2\npyyaml==6.0.2\npyzmq==26.2.0\nrequests==2.32.3\nrich==13.8.1\nsetuptools==80.9.0\nshimmy==1.3.0\nshtab==1.7.2\nsix==1.16.0\nstable-baselines3==2.3.2\nstack-data==0.6.3\nswig==4.2.1\nsympy==1.13.2\ntensorboard==2.17.1\ntensorboard-data-server==0.7.2\ntomli==2.0.1\ntorch==2.4.1\ntornado==6.4.1\ntqdm==4.66.5\ntraitlets==5.14.3\ntriton==3.0.0\ntypeguard==4.4.4\ntyping-extensions==4.12.2\ntyro==0.9.32\ntzdata==2024.1\nurllib3==2.2.2\nvizdoom==1.2.4\nwcwidth==0.2.13\nwerkzeug==3.0.4\nwidgetsnbextension==4.0.13\nxxhash==3.5.0\nyarl==1.11.1\nzipp==3.20.1\n",pip-requirements,tab
+18,83408,"TERMINAL",0,0,"",,terminal_command
+19,205566,"TERMINAL",0,0,"bash",,terminal_focus
+20,231933,"TERMINAL",0,0,"cd data/",,terminal_command
+21,236765,"TERMINAL",0,0,"source .venv/bin/activate",,terminal_command
+22,241715,"TERMINAL",0,0,"python --version",,terminal_command
+23,241771,"TERMINAL",0,0,"]633;CPython 3.10.18\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+24,246514,"TERMINAL",0,0,"bash",,terminal_focus
+25,455668,"TERMINAL",0,0,"bash",,terminal_focus
+26,459770,"TERMINAL",0,0,"swig --version",,terminal_command
+27,459830,"TERMINAL",0,0,"]633;C",,terminal_output
+28,460705,"TERMINAL",0,0,"Unable to find option or file '--version', Use 'swig -help' for more information.\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+29,1074247,"TERMINAL",0,0,"ls",,terminal_command
+30,1074257,"TERMINAL",0,0,"]633;Cjasmine_data  pyproject.toml  requirements_doom.txt  uv.lock  _vizdoom  _vizdoom.ini\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+31,1088085,"TERMINAL",0,0,"ls data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",,terminal_command
+32,1088119,"TERMINAL",0,0,"]633;Cls: cannot access 'data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py': No such file or directory\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+33,1097483,"TERMINAL",0,0,"ls .venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",,terminal_command
+34,1097487,"TERMINAL",0,0,"]633;C.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py\r\n]0;tum_cte0515@hkn1993:~/Projects/jasmine/data",,terminal_output
+35,1099456,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",0,0,"""""""\nSave util taken from stable_baselines\nused to serialize data (class parameters) of model classes\n""""""\n\nimport base64\nimport functools\nimport io\nimport json\nimport os\nimport pathlib\nimport pickle\nimport warnings\nimport zipfile\nfrom typing import Any, Dict, Optional, Tuple, Union\n\nimport cloudpickle\nimport torch as th\n\nimport stable_baselines3 as sb3\nfrom stable_baselines3.common.type_aliases import TensorDict\nfrom stable_baselines3.common.utils import get_device, get_system_info\n\n\ndef recursive_getattr(obj: Any, attr: str, *args) -> Any:\n    """"""\n    Recursive version of getattr\n    taken from https://stackoverflow.com/questions/31174295\n\n    Ex:\n    > MyObject.sub_object = SubObject(name='test')\n    > recursive_getattr(MyObject, 'sub_object.name')  # return test\n    :param obj:\n    :param attr: Attribute to retrieve\n    :return: The attribute\n    """"""\n\n    def _getattr(obj: Any, attr: str) -> Any:\n        return getattr(obj, attr, *args)\n\n    return functools.reduce(_getattr, [obj, *attr.split(""."")])\n\n\ndef recursive_setattr(obj: Any, attr: str, val: Any) -> None:\n    """"""\n    Recursive version of setattr\n    taken from https://stackoverflow.com/questions/31174295\n\n    Ex:\n    > MyObject.sub_object = SubObject(name='test')\n    > recursive_setattr(MyObject, 'sub_object.name', 'hello')\n    :param obj:\n    :param attr: Attribute to set\n    :param val: New value of the attribute\n    """"""\n    pre, _, post = attr.rpartition(""."")\n    return setattr(recursive_getattr(obj, pre) if pre else obj, post, val)\n\n\ndef is_json_serializable(item: Any) -> bool:\n    """"""\n    Test if an object is serializable into JSON\n\n    :param item: The object to be tested for JSON serialization.\n    :return: True if object is JSON serializable, false otherwise.\n    """"""\n    # Try with try-except struct.\n    json_serializable = True\n    try:\n        _ = json.dumps(item)\n    except TypeError:\n        json_serializable = False\n    return json_serializable\n\n\ndef data_to_json(data: Dict[str, Any]) -> str:\n    """"""\n    Turn data (class parameters) into a JSON string for storing\n\n    :param data: Dictionary of class parameters to be\n        stored. Items that are not JSON serializable will be\n        pickled with Cloudpickle and stored as bytearray in\n        the JSON file\n    :return: JSON string of the data serialized.\n    """"""\n    # First, check what elements can not be JSONfied,\n    # and turn them into byte-strings\n    serializable_data = {}\n    for data_key, data_item in data.items():\n        # See if object is JSON serializable\n        if is_json_serializable(data_item):\n            # All good, store as it is\n            serializable_data[data_key] = data_item\n        else:\n            # Not serializable, cloudpickle it into\n            # bytes and convert to base64 string for storing.\n            # Also store type of the class for consumption\n            # from other languages/humans, so we have an\n            # idea what was being stored.\n            base64_encoded = base64.b64encode(cloudpickle.dumps(data_item)).decode()\n\n            # Use "":"" to make sure we do\n            # not override these keys\n            # when we include variables of the object later\n            cloudpickle_serialization = {\n                "":type:"": str(type(data_item)),\n                "":serialized:"": base64_encoded,\n            }\n\n            # Add first-level JSON-serializable items of the\n            # object for further details (but not deeper than this to\n            # avoid deep nesting).\n            # First we check that object has attributes (not all do,\n            # e.g. numpy scalars)\n            if hasattr(data_item, ""__dict__"") or isinstance(data_item, dict):\n                # Take elements from __dict__ for custom classes\n                item_generator = data_item.items if isinstance(data_item, dict) else data_item.__dict__.items\n                for variable_name, variable_item in item_generator():\n                    # Check if serializable. If not, just include the\n                    # string-representation of the object.\n                    if is_json_serializable(variable_item):\n                        cloudpickle_serialization[variable_name] = variable_item\n                    else:\n                        cloudpickle_serialization[variable_name] = str(variable_item)\n\n            serializable_data[data_key] = cloudpickle_serialization\n    json_string = json.dumps(serializable_data, indent=4)\n    return json_string\n\n\ndef json_to_data(json_string: str, custom_objects: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:\n    """"""\n    Turn JSON serialization of class-parameters back into dictionary.\n\n    :param json_string: JSON serialization of the class-parameters\n        that should be loaded.\n    :param custom_objects: Dictionary of objects to replace\n        upon loading. If a variable is present in this dictionary as a\n        key, it will not be deserialized and the corresponding item\n        will be used instead. Similar to custom_objects in\n        ``keras.models.load_model``. Useful when you have an object in\n        file that can not be deserialized.\n    :return: Loaded class parameters.\n    """"""\n    if custom_objects is not None and not isinstance(custom_objects, dict):\n        raise ValueError(""custom_objects argument must be a dict or None"")\n\n    json_dict = json.loads(json_string)\n    # This will be filled with deserialized data\n    return_data = {}\n    for data_key, data_item in json_dict.items():\n        if custom_objects is not None and data_key in custom_objects.keys():\n            # If item is provided in custom_objects, replace\n            # the one from JSON with the one in custom_objects\n            return_data[data_key] = custom_objects[data_key]\n        elif isinstance(data_item, dict) and "":serialized:"" in data_item.keys():\n            # If item is dictionary with "":serialized:""\n            # key, this means it is serialized with cloudpickle.\n            serialization = data_item["":serialized:""]\n            # Try-except deserialization in case we run into\n            # errors. If so, we can tell bit more information to\n            # user.\n            try:\n                base64_object = base64.b64decode(serialization.encode())\n                deserialized_object = cloudpickle.loads(base64_object)\n            except (RuntimeError, TypeError, AttributeError) as e:\n                warnings.warn(\n                    f""Could not deserialize object {data_key}. ""\n                    ""Consider using `custom_objects` argument to replace ""\n                    ""this object.\n""\n                    f""Exception: {e}""\n                )\n            else:\n                return_data[data_key] = deserialized_object\n        else:\n            # Read as it is\n            return_data[data_key] = data_item\n    return return_data\n\n\[email protected]\ndef open_path(\n    path: Union[str, pathlib.Path, io.BufferedIOBase], mode: str, verbose: int = 0, suffix: Optional[str] = None\n) -> Union[io.BufferedWriter, io.BufferedReader, io.BytesIO]:\n    """"""\n    Opens a path for reading or writing with a preferred suffix and raises debug information.\n    If the provided path is a derivative of io.BufferedIOBase it ensures that the file\n    matches the provided mode, i.e. If the mode is read (""r"", ""read"") it checks that the path is readable.\n    If the mode is write (""w"", ""write"") it checks that the file is writable.\n\n    If the provided path is a string or a pathlib.Path, it ensures that it exists. If the mode is ""read""\n    it checks that it exists, if it doesn't exist it attempts to read path.suffix if a suffix is provided.\n    If the mode is ""write"" and the path does not exist, it creates all the parent folders. If the path\n    points to a folder, it changes the path to path_2. If the path already exists and verbose >= 2,\n    it raises a warning.\n\n    :param path: the path to open.\n        if save_path is a str or pathlib.Path and mode is ""w"", single dispatch ensures that the\n        path actually exists. If path is a io.BufferedIOBase the path exists.\n    :param mode: how to open the file. ""w""|""write"" for writing, ""r""|""read"" for reading.\n    :param verbose: Verbosity level: 0 for no output, 1 for info messages, 2 for debug messages\n    :param suffix: The preferred suffix. If mode is ""w"" then the opened file has the suffix.\n        If mode is ""r"" then we attempt to open the path. If an error is raised and the suffix\n        is not None, we attempt to open the path with the suffix.\n    :return:\n    """"""\n    # Note(antonin): the true annotation should be IO[bytes]\n    # but there is not easy way to check that\n    allowed_types = (io.BufferedWriter, io.BufferedReader, io.BytesIO, io.BufferedRandom)\n    if not isinstance(path, allowed_types):\n        raise TypeError(f""Path {path} parameter has invalid type: expected one of {allowed_types}."")\n    if path.closed:\n        raise ValueError(f""File stream {path} is closed."")\n    mode = mode.lower()\n    try:\n        mode = {""write"": ""w"", ""read"": ""r"", ""w"": ""w"", ""r"": ""r""}[mode]\n    except KeyError as e:\n        raise ValueError(""Expected mode to be either 'w' or 'r'."") from e\n    if (""w"" == mode) and not path.writable() or (""r"" == mode) and not path.readable():\n        error_msg = ""writable"" if ""w"" == mode else ""readable""\n        raise ValueError(f""Expected a {error_msg} file."")\n    return path\n\n\n@open_path.register(str)\ndef open_path_str(path: str, mode: str, verbose: int = 0, suffix: Optional[str] = None) -> io.BufferedIOBase:\n    """"""\n    Open a path given by a string. If writing to the path, the function ensures\n    that the path exists.\n\n    :param path: the path to open. If mode is ""w"" then it ensures that the path exists\n        by creating the necessary folders and renaming path if it points to a folder.\n    :param mode: how to open the file. ""w"" for writing, ""r"" for reading.\n    :param verbose: Verbosity level: 0 for no output, 1 for info messages, 2 for debug messages\n    :param suffix: The preferred suffix. If mode is ""w"" then the opened file has the suffix.\n        If mode is ""r"" then we attempt to open the path. If an error is raised and the suffix\n        is not None, we attempt to open the path with the suffix.\n    :return:\n    """"""\n    return open_path_pathlib(pathlib.Path(path), mode, verbose, suffix)\n\n\n@open_path.register(pathlib.Path)\ndef open_path_pathlib(path: pathlib.Path, mode: str, verbose: int = 0, suffix: Optional[str] = None) -> io.BufferedIOBase:\n    """"""\n    Open a path given by a string. If writing to the path, the function ensures\n    that the path exists.\n\n    :param path: the path to check. If mode is ""w"" then it\n        ensures that the path exists by creating the necessary folders and\n        renaming path if it points to a folder.\n    :param mode: how to open the file. ""w"" for writing, ""r"" for reading.\n    :param verbose: Verbosity level: 0 for no output, 2 for indicating if path without suffix is not found when mode is ""r""\n    :param suffix: The preferred suffix. If mode is ""w"" then the opened file has the suffix.\n        If mode is ""r"" then we attempt to open the path. If an error is raised and the suffix\n        is not None, we attempt to open the path with the suffix.\n    :return:\n    """"""\n    if mode not in (""w"", ""r""):\n        raise ValueError(""Expected mode to be either 'w' or 'r'."")\n\n    if mode == ""r"":\n        try:\n            return open_path(path.open(""rb""), mode, verbose, suffix)\n        except FileNotFoundError as error:\n            if suffix is not None and suffix != """":\n                newpath = pathlib.Path(f""{path}.{suffix}"")\n                if verbose >= 2:\n                    warnings.warn(f""Path '{path}' not found. Attempting {newpath}."")\n                path, suffix = newpath, None\n            else:\n                raise error\n    else:\n        try:\n            if path.suffix == """" and suffix is not None and suffix != """":\n                path = pathlib.Path(f""{path}.{suffix}"")\n            if path.exists() and path.is_file() and verbose >= 2:\n                warnings.warn(f""Path '{path}' exists, will overwrite it."")\n            return open_path(path.open(""wb""), mode, verbose, suffix)\n        except IsADirectoryError:\n            warnings.warn(f""Path '{path}' is a folder. Will save instead to {path}_2"")\n            path = pathlib.Path(f""{path}_2"")\n        except FileNotFoundError:  # Occurs when the parent folder doesn't exist\n            warnings.warn(f""Path '{path.parent}' does not exist. Will create it."")\n            path.parent.mkdir(exist_ok=True, parents=True)\n\n    # if opening was successful uses the open_path() function\n    # if opening failed with IsADirectory|FileNotFound, calls open_path_pathlib\n    #   with corrections\n    # if reading failed with FileNotFoundError, calls open_path_pathlib with suffix\n    return open_path_pathlib(path, mode, verbose, suffix)\n\n\ndef save_to_zip_file(\n    save_path: Union[str, pathlib.Path, io.BufferedIOBase],\n    data: Optional[Dict[str, Any]] = None,\n    params: Optional[Dict[str, Any]] = None,\n    pytorch_variables: Optional[Dict[str, Any]] = None,\n    verbose: int = 0,\n) -> None:\n    """"""\n    Save model data to a zip archive.\n\n    :param save_path: Where to store the model.\n        if save_path is a str or pathlib.Path ensures that the path actually exists.\n    :param data: Class parameters being stored (non-PyTorch variables)\n    :param params: Model parameters being stored expected to contain an entry for every\n                   state_dict with its name and the state_dict.\n    :param pytorch_variables: Other PyTorch variables expected to contain name and value of the variable.\n    :param verbose: Verbosity level: 0 for no output, 1 for info messages, 2 for debug messages\n    """"""\n    file = open_path(save_path, ""w"", verbose=0, suffix=""zip"")\n    # data/params can be None, so do not\n    # try to serialize them blindly\n    if data is not None:\n        serialized_data = data_to_json(data)\n\n    # Create a zip-archive and write our objects there.\n    with zipfile.ZipFile(file, mode=""w"") as archive:\n        # Do not try to save ""None"" elements\n        if data is not None:\n            archive.writestr(""data"", serialized_data)\n        if pytorch_variables is not None:\n            with archive.open(""pytorch_variables.pth"", mode=""w"", force_zip64=True) as pytorch_variables_file:\n                th.save(pytorch_variables, pytorch_variables_file)\n        if params is not None:\n            for file_name, dict_ in params.items():\n                with archive.open(file_name + "".pth"", mode=""w"", force_zip64=True) as param_file:\n                    th.save(dict_, param_file)\n        # Save metadata: library version when file was saved\n        archive.writestr(""_stable_baselines3_version"", sb3.__version__)\n        # Save system info about the current python env\n        archive.writestr(""system_info.txt"", get_system_info(print_info=False)[1])\n\n    if isinstance(save_path, (str, pathlib.Path)):\n        file.close()\n\n\ndef save_to_pkl(path: Union[str, pathlib.Path, io.BufferedIOBase], obj: Any, verbose: int = 0) -> None:\n    """"""\n    Save an object to path creating the necessary folders along the way.\n    If the path exists and is a directory, it will raise a warning and rename the path.\n    If a suffix is provided in the path, it will use that suffix, otherwise, it will use '.pkl'.\n\n    :param path: the path to open.\n        if save_path is a str or pathlib.Path and mode is ""w"", single dispatch ensures that the\n        path actually exists. If path is a io.BufferedIOBase the path exists.\n    :param obj: The object to save.\n    :param verbose: Verbosity level: 0 for no output, 1 for info messages, 2 for debug messages\n    """"""\n    file = open_path(path, ""w"", verbose=verbose, suffix=""pkl"")\n    # Use protocol>=4 to support saving replay buffers >= 4Gb\n    # See https://docs.python.org/3/library/pickle.html\n    pickle.dump(obj, file, protocol=pickle.HIGHEST_PROTOCOL)\n    if isinstance(path, (str, pathlib.Path)):\n        file.close()\n\n\ndef load_from_pkl(path: Union[str, pathlib.Path, io.BufferedIOBase], verbose: int = 0) -> Any:\n    """"""\n    Load an object from the path. If a suffix is provided in the path, it will use that suffix.\n    If the path does not exist, it will attempt to load using the .pkl suffix.\n\n    :param path: the path to open.\n        if save_path is a str or pathlib.Path and mode is ""w"", single dispatch ensures that the\n        path actually exists. If path is a io.BufferedIOBase the path exists.\n    :param verbose: Verbosity level: 0 for no output, 1 for info messages, 2 for debug messages\n    """"""\n    file = open_path(path, ""r"", verbose=verbose, suffix=""pkl"")\n    obj = pickle.load(file)\n    if isinstance(path, (str, pathlib.Path)):\n        file.close()\n    return obj\n\n\ndef load_from_zip_file(\n    load_path: Union[str, pathlib.Path, io.BufferedIOBase],\n    load_data: bool = True,\n    custom_objects: Optional[Dict[str, Any]] = None,\n    device: Union[th.device, str] = ""auto"",\n    verbose: int = 0,\n    print_system_info: bool = False,\n) -> Tuple[Optional[Dict[str, Any]], TensorDict, Optional[TensorDict]]:\n    """"""\n    Load model data from a .zip archive\n\n    :param load_path: Where to load the model from\n    :param load_data: Whether we should load and return data\n        (class parameters). Mainly used by 'load_parameters' to only load model parameters (weights)\n    :param custom_objects: Dictionary of objects to replace\n        upon loading. If a variable is present in this dictionary as a\n        key, it will not be deserialized and the corresponding item\n        will be used instead. Similar to custom_objects in\n        ``keras.models.load_model``. Useful when you have an object in\n        file that can not be deserialized.\n    :param device: Device on which the code should run.\n    :param verbose: Verbosity level: 0 for no output, 1 for info messages, 2 for debug messages\n    :param print_system_info: Whether to print or not the system info\n        about the saved model.\n    :return: Class parameters, model state_dicts (aka ""params"", dict of state_dict)\n        and dict of pytorch variables\n    """"""\n    file = open_path(load_path, ""r"", verbose=verbose, suffix=""zip"")\n\n    # set device to cpu if cuda is not available\n    device = get_device(device=device)\n\n    # Open the zip archive and load data\n    try:\n        with zipfile.ZipFile(file) as archive:\n            namelist = archive.namelist()\n            # If data or parameters is not in the\n            # zip archive, assume they were stored\n            # as None (_save_to_file_zip allows this).\n            data = None\n            pytorch_variables = None\n            params = {}\n\n            # Debug system info first\n            if print_system_info:\n                if ""system_info.txt"" in namelist:\n                    print(""== SAVED MODEL SYSTEM INFO =="")\n                    print(archive.read(""system_info.txt"").decode())\n                else:\n                    warnings.warn(\n                        ""The model was saved with SB3 <= 1.2.0 and thus cannot print system information."",\n                        UserWarning,\n                    )\n\n            if ""data"" in namelist and load_data:\n                # Load class parameters that are stored\n                # with either JSON or pickle (not PyTorch variables).\n                json_data = archive.read(""data"").decode()\n                data = json_to_data(json_data, custom_objects=custom_objects)\n\n            # Check for all .pth files and load them using th.load.\n            # ""pytorch_variables.pth"" stores PyTorch variables, and any other .pth\n            # files store state_dicts of variables with custom names (e.g. policy, policy.optimizer)\n            pth_files = [file_name for file_name in namelist if os.path.splitext(file_name)[1] == "".pth""]\n            for file_path in pth_files:\n                with archive.open(file_path, mode=""r"") as param_file:\n                    # File has to be seekable, but param_file is not, so load in BytesIO first\n                    # fixed in python >= 3.7\n                    file_content = io.BytesIO()\n                    file_content.write(param_file.read())\n                    # go to start of file\n                    file_content.seek(0)\n                    # Load the parameters with the right ``map_location``.\n                    # Remove "".pth"" ending with splitext\n                    # Note(antonin): we cannot use weights_only=True, as it breaks with PyTorch 1.13, see GH#1911\n                    th_object = th.load(file_content, map_location=device, weights_only=False)\n                    # ""tensors.pth"" was renamed ""pytorch_variables.pth"" in v0.9.0, see PR #138\n                    if file_path == ""pytorch_variables.pth"" or file_path == ""tensors.pth"":\n                        # PyTorch variables (not state_dicts)\n                        pytorch_variables = th_object\n                    else:\n                        # State dicts. Store into params dictionary\n                        # with same name as in .zip file (without .pth)\n                        params[os.path.splitext(file_path)[0]] = th_object\n    except zipfile.BadZipFile as e:\n        # load_path wasn't a zip file\n        raise ValueError(f""Error: the file {load_path} wasn't a zip-file"") from e\n    finally:\n        if isinstance(load_path, (str, pathlib.Path)):\n            file.close()\n    return data, params, pytorch_variables\n",python,tab
+36,1102104,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",4188,0,"",python,selection_mouse
+37,1108461,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",5092,0,"",python,selection_mouse
+38,1112406,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6294,0,"",python,selection_mouse
+39,1112418,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6293,0,"",python,selection_command
+40,1112650,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6293,1,")",python,selection_mouse
+41,1112652,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6258,35,"t = cloudpickle.loads(base64_object",python,selection_mouse
+42,1112654,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6255,38,"ject = cloudpickle.loads(base64_object",python,selection_mouse
+43,1112657,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6294,0,"",python,selection_command
+44,1112709,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6253,41,"object = cloudpickle.loads(base64_object)",python,selection_mouse
+45,1112710,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6251,43,"d_object = cloudpickle.loads(base64_object)",python,selection_mouse
+46,1112711,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6250,44,"ed_object = cloudpickle.loads(base64_object)",python,selection_mouse
+47,1112717,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6248,46,"ized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+48,1112751,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6246,48,"alized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+49,1112767,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6245,49,"ialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+50,1112794,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6244,50,"rialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+51,1112795,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6243,51,"erialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+52,1112833,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6242,52,"serialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+53,1112847,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6241,53,"eserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+54,1112882,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6240,54,"deserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+55,1112883,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6239,55," deserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+56,1112911,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6238,56,"  deserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+57,1112933,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6237,57,"   deserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+58,1112962,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6236,58,"    deserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+59,1112994,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6235,59,"     deserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+60,1113047,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6234,60,"      deserialized_object = cloudpickle.loads(base64_object)",python,selection_mouse
+61,1113401,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6234,0,"",python,selection_mouse
+62,1113504,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6224,16,"                ",python,selection_mouse
+63,1114025,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6234,0,"",python,selection_mouse
+64,1114086,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6224,16,"                ",python,selection_mouse
+65,1114236,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6224,71,"                deserialized_object = cloudpickle.loads(base64_object)\n",python,selection_mouse
+66,1115610,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6244,0,"",python,selection_mouse
+67,1115689,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6240,19,"deserialized_object",python,selection_mouse
+68,1115819,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6224,71,"                deserialized_object = cloudpickle.loads(base64_object)\n",python,selection_mouse
+69,1134605,"data/.venv/lib/python3.10/site-packages/stable_baselines3/common/save_util.py",6499,0,"",python,selection_mouse

af40c12958422f63ff638ecf057ca5960a6e79dafb430c2e4343b991cc9cefcc/crowd-code-ee716c9f-2894-4e4a-83fd-1afc9628a5fd1760867766844-2025_10_19-11.57.29.153/source.csv

@@ -0,0 +1,203 @@
+Sequence,Time,File,RangeOffset,RangeLength,Text,Language,Type
+1,5,"jasmine/train_tokenizer.py",0,0,"import os\n\nos.environ.setdefault(""XLA_PYTHON_CLIENT_MEM_FRACTION"", ""0.98"")\n\nfrom dataclasses import dataclass, field\nfrom typing import cast, Optional\n\nimport einops\nimport itertools\nfrom jax.sharding import Mesh, PartitionSpec, NamedSharding\nfrom jax.experimental.mesh_utils import create_device_mesh\nimport optax\nimport orbax.checkpoint as ocp\nimport numpy as np\nimport dm_pix as pix\nimport jax\nimport jax.numpy as jnp\nimport tyro\nimport wandb\nimport grain\nimport flax.nnx as nnx\n\nfrom models.tokenizer import TokenizerVQVAE\nfrom utils.dataloader import get_dataloader\nfrom utils.train_utils import (\n    get_lr_schedule,\n    count_parameters_by_component,\n    print_mem_stats,\n    print_compiled_memory_stats,\n    print_compiled_cost_analysis,\n)\n\n\n@dataclass\nclass Args:\n    # Experiment\n    num_steps: int = 300_000\n    seed: int = 0\n    seq_len: int = 16\n    image_channels: int = 3\n    image_height: int = 64\n    image_width: int = 64\n    data_dir: str = """"\n    save_ckpt: bool = False\n    restore_ckpt: bool = False\n    # Optimization\n    vq_beta: float = 0.25\n    batch_size: int = 48\n    init_lr: float = 0.0\n    max_lr: float = 3e-4\n    decay_end: float = 0.0\n    wsd_decay_steps: int = (\n        30_000  # NOTE: wsd_decay_steps will only be used when using a wsd-schedule\n    )\n    lr_schedule: str = ""wsd""  # supported options: wsd, cos\n    warmup_steps: int = 10000\n    # Tokenizer\n    model_dim: int = 512\n    ffn_dim: int = 2048\n    latent_dim: int = 32\n    num_latents: int = 1024\n    patch_size: int = 16\n    num_blocks: int = 4\n    num_heads: int = 8\n    dropout: float = 0.0\n    codebook_dropout: float = 0.01\n    param_dtype = jnp.float32\n    dtype = jnp.bfloat16\n    use_flash_attention: bool = True\n    # Logging\n    log: bool = True\n    entity: str = """"\n    project: str = """"\n    name: str = ""train_tokenizer""\n    tags: list[str] = field(default_factory=lambda: [""tokenizer""])\n    log_interval: int = 50\n    log_image_interval: int = 1000\n    ckpt_dir: str = """"\n    log_checkpoint_interval: int = 1000\n    log_checkpoint_keep_period: int = 20_000\n    log_gradients: bool = False\n    val_data_dir: str = """"\n    val_interval: int = 20_000\n    val_steps: int = 50\n    wandb_id: str = """"\n\n\ndef build_model(args: Args, rng: jax.Array) -> tuple[TokenizerVQVAE, jax.Array]:\n    rng, _rng = jax.random.split(rng)\n    rngs = nnx.Rngs(_rng)\n    return (\n        TokenizerVQVAE(\n            in_dim=args.image_channels,\n            model_dim=args.model_dim,\n            ffn_dim=args.ffn_dim,\n            latent_dim=args.latent_dim,\n            num_latents=args.num_latents,\n            patch_size=args.patch_size,\n            num_blocks=args.num_blocks,\n            num_heads=args.num_heads,\n            dropout=args.dropout,\n            codebook_dropout=args.codebook_dropout,\n            param_dtype=args.param_dtype,\n            dtype=args.dtype,\n            use_flash_attention=args.use_flash_attention,\n            rngs=rngs,\n        ),\n        rng,\n    )\n\n\ndef build_optimizer(model: TokenizerVQVAE, args: Args) -> nnx.ModelAndOptimizer:\n    lr_schedule = get_lr_schedule(\n        args.lr_schedule,\n        args.init_lr,\n        args.max_lr,\n        args.decay_end,\n        args.num_steps,\n        args.warmup_steps,\n        args.wsd_decay_steps,\n    )\n    tx = optax.adamw(\n        learning_rate=lr_schedule,\n        b1=0.9,\n        b2=0.9,\n        weight_decay=1e-4,\n        mu_dtype=args.param_dtype,  # moments in full precision\n    )\n    optimizer = nnx.ModelAndOptimizer(model, tx)\n    return optimizer\n\n\ndef build_mesh_and_sharding(\n    num_devices: int,\n) -> tuple[Mesh, NamedSharding, NamedSharding]:\n    device_mesh_arr = create_device_mesh((num_devices,))\n    mesh = Mesh(devices=device_mesh_arr, axis_names=(""data"",))\n    replicated_sharding = NamedSharding(mesh, PartitionSpec())\n    videos_sharding = NamedSharding(mesh, PartitionSpec(""data"", None, None, None, None))\n    return mesh, replicated_sharding, videos_sharding\n\n\ndef shard_optimizer_states(\n    optimizer: nnx.ModelAndOptimizer, replicated_sharding: NamedSharding\n) -> None:\n    model_state = nnx.state(optimizer.model)\n    model_sharded_state = jax.lax.with_sharding_constraint(\n        model_state, replicated_sharding\n    )\n    nnx.update(optimizer.model, model_sharded_state)\n    optimizer_state = nnx.state(optimizer, nnx.optimizer.OptState)\n    optimizer_sharded_state = jax.lax.with_sharding_constraint(\n        optimizer_state, replicated_sharding\n    )\n    nnx.update(optimizer, optimizer_sharded_state)\n\n\ndef build_dataloader(args: Args, data_dir: str) -> grain.DataLoaderIterator:\n    image_shape = (args.image_height, args.image_width, args.image_channels)\n    array_record_files = [\n        os.path.join(data_dir, x)\n        for x in os.listdir(data_dir)\n        if x.endswith("".array_record"")\n    ]\n    grain_dataloader = get_dataloader(\n        array_record_files,\n        args.seq_len,\n        # NOTE: We deliberately pass the global batch size\n        # The dataloader shards the dataset across all processes\n        args.batch_size,\n        *image_shape,\n        num_workers=8,\n        prefetch_buffer_size=1,\n        seed=args.seed,\n    )\n    initial_state = grain_dataloader._create_initial_state()\n    grain_iterator = grain.DataLoaderIterator(grain_dataloader, initial_state)\n    return grain_iterator\n\n\ndef build_checkpoint_manager(args: Args) -> Optional[ocp.CheckpointManager]:\n    if args.restore_ckpt or args.save_ckpt:\n        handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n        handler_registry.add(\n            ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n        )\n        handler_registry.add(\n            ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n        )\n        handler_registry.add(\n            ""train_dataloader_state"",\n            grain.checkpoint.CheckpointSave,\n            cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n        )\n        handler_registry.add(\n            ""train_dataloader_state"",\n            grain.checkpoint.CheckpointRestore,\n            cast(ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler),\n        )\n        if args.val_data_dir:\n            handler_registry.add(\n                ""val_dataloader_state"",\n                grain.checkpoint.CheckpointSave,\n                cast(\n                    ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler\n                ),\n            )\n            handler_registry.add(\n                ""val_dataloader_state"",\n                grain.checkpoint.CheckpointRestore,\n                cast(\n                    ocp.handlers.CheckpointHandler, grain.checkpoint.CheckpointHandler\n                ),\n            )\n        checkpoint_options = ocp.CheckpointManagerOptions(\n            save_interval_steps=args.log_checkpoint_interval,\n            max_to_keep=3,\n            keep_period=args.log_checkpoint_keep_period,\n            step_format_fixed_length=6,\n            cleanup_tmp_directories=True,\n        )\n        checkpoint_manager = ocp.CheckpointManager(\n            args.ckpt_dir,\n            options=checkpoint_options,\n            handler_registry=handler_registry,\n        )\n        return checkpoint_manager\n    else:\n        return None\n\n\ndef restore_checkpoint_if_needed(\n    args: Args,\n    checkpoint_manager: Optional[ocp.CheckpointManager],\n    optimizer: nnx.ModelAndOptimizer,\n    train_iterator: grain.DataLoaderIterator,\n    val_iterator: Optional[grain.DataLoaderIterator],\n    restore_step: Optional[int] = None,\n) -> tuple[\n    int, nnx.ModelAndOptimizer, grain.DataLoaderIterator, grain.DataLoaderIterator\n]:\n    step = 0\n    if checkpoint_manager and restore_step is None:\n        restore_step = checkpoint_manager.latest_step()\n    if args.restore_ckpt:\n        assert checkpoint_manager is not None\n        abstract_optimizer = nnx.eval_shape(lambda: optimizer)\n        abstract_optimizer_state = nnx.state(abstract_optimizer)\n        if val_iterator:\n            restore_args = ocp.args.Composite(\n                model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n                train_dataloader_state=grain.checkpoint.CheckpointRestore(train_iterator),  # type: ignore\n                val_dataloader_state=grain.checkpoint.CheckpointRestore(val_iterator),  # type: ignore\n            )\n        else:\n            restore_args = ocp.args.Composite(\n                model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n                train_dataloader_state=grain.checkpoint.CheckpointRestore(train_iterator),  # type: ignore\n            )\n        restored = checkpoint_manager.restore(restore_step, args=restore_args)\n        restored_optimizer_state = restored[""model_state""]\n        nnx.update(optimizer, restored_optimizer_state)\n        train_iterator = restored[""train_dataloader_state""]\n        if val_iterator:\n            val_iterator = restored[""val_dataloader_state""]\n        step = restore_step or 0\n        print(f""Restored dataloader and model state from step {step}"")\n    return step, optimizer, train_iterator, val_iterator\n\n\ndef main(args: Args) -> None:\n    jax.distributed.initialize()\n    num_devices = jax.device_count()\n    if num_devices == 0:\n        raise ValueError(""No JAX devices found."")\n    print(f""Running on {num_devices} devices."")\n\n    if args.batch_size % num_devices != 0:\n        raise ValueError(\n            f""Global batch size {args.batch_size} must be divisible by ""\n            f""number of devices {num_devices}.""\n        )\n\n    rng = jax.random.key(args.seed)\n\n    # --- Initialize model ---\n    tokenizer, rng = build_model(args, rng)\n\n    _, params, _ = nnx.split(tokenizer, nnx.Param, ...)\n    param_counts = count_parameters_by_component(params)\n\n    if args.log and jax.process_index() == 0:\n        wandb_init_kwargs = {\n            ""entity"": args.entity,\n            ""project"": args.project,\n            ""name"": args.name,\n            ""tags"": args.tags,\n            ""group"": ""debug"",\n            ""config"": args,\n        }\n\n        if args.wandb_id:\n            wandb_init_kwargs.update(\n                {\n                    ""id"": args.wandb_id,\n                    ""resume"": ""allow"",\n                }\n            )\n        wandb.init(**wandb_init_kwargs)\n\n        wandb.config.update({""model_param_count"": param_counts})\n\n    print(""Parameter counts:"")\n    print(param_counts)\n\n    # --- Initialize optimizer ---\n    optimizer = build_optimizer(tokenizer, args)\n    del tokenizer\n\n    # FIXME: switch to create_hybrid_device_mesh for runs spanning multiple nodes\n    _, replicated_sharding, videos_sharding = build_mesh_and_sharding(num_devices)\n\n    shard_optimizer_states(optimizer, replicated_sharding)\n\n    # --- Initialize checkpoint manager ---\n    checkpoint_manager = build_checkpoint_manager(args)\n\n    # --- Create DataLoaderIterator from dataloader ---\n    train_iterator = build_dataloader(args, args.data_dir)\n    val_iterator = None\n    if args.val_data_dir:\n        val_iterator = build_dataloader(args, args.val_data_dir)\n\n    # --- Restore checkpoint ---\n    step, optimizer, train_iterator, val_iterator = restore_checkpoint_if_needed(\n        args, checkpoint_manager, optimizer, train_iterator, val_iterator\n    )\n\n    # --- Define loss and train step (close over args) ---\n    def tokenizer_loss_fn(\n        model: TokenizerVQVAE, inputs: dict, training: bool = False\n    ) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n        gt = jnp.asarray(inputs[""videos""], dtype=jnp.float32) / 255.0\n        inputs[""videos""] = gt.astype(args.dtype)\n        outputs = model(inputs, training=training)\n        outputs[""recon""] = outputs[""recon""].astype(jnp.float32)\n        mse = jnp.square(gt - outputs[""recon""]).mean()\n        q_loss = jnp.square(jax.lax.stop_gradient(outputs[""emb""]) - outputs[""z""]).mean()\n        commitment_loss = jnp.square(\n            outputs[""emb""] - jax.lax.stop_gradient(outputs[""z""])\n        ).mean()\n        loss = mse + q_loss + args.vq_beta * commitment_loss\n\n        gt_clipped = gt.clip(0, 1).reshape(-1, *gt.shape[2:])\n        recon = outputs[""recon""].clip(0, 1).reshape(-1, *outputs[""recon""].shape[2:])\n        psnr = jnp.asarray(pix.psnr(gt_clipped, recon)).mean()\n        ssim = jnp.asarray(pix.ssim(gt_clipped, recon)).mean()\n        _, index_counts = jnp.unique_counts(\n            jnp.ravel(outputs[""indices""]), size=args.num_latents, fill_value=0\n        )\n        codebook_usage = (index_counts != 0).mean()\n        metrics = dict(\n            loss=loss,\n            mse=mse,\n            q_loss=q_loss,\n            commitment_loss=commitment_loss,\n            psnr=psnr,\n            ssim=ssim,\n            codebook_usage=codebook_usage,\n        )\n        return loss, (outputs[""recon""], metrics)\n\n    @nnx.jit(donate_argnums=0)\n    def train_step(\n        optimizer: nnx.ModelAndOptimizer, inputs: dict\n    ) -> tuple[jax.Array, jax.Array, dict]:\n        def loss_fn(model: TokenizerVQVAE) -> tuple[jax.Array, tuple[jax.Array, dict]]:\n            model.train()\n            return tokenizer_loss_fn(model, inputs, training=True)\n\n        (loss, (recon, metrics)), grads = nnx.value_and_grad(loss_fn, has_aux=True)(\n            optimizer.model\n        )\n        optimizer.update(grads)\n        if args.log_gradients:\n            metrics[""encoder_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""encoder""]\n            )\n            metrics[""vq_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""vq""]\n            )\n            metrics[""decoder_gradients_std/""] = jax.tree.map(\n                lambda x: x.std(), grads[""params""][""decoder""]\n            )\n        return loss, recon, metrics\n\n    @nnx.jit\n    def val_step(\n        tokenizer: TokenizerVQVAE, inputs: dict\n    ) -> tuple[jax.Array, jax.Array, dict]:\n        tokenizer.eval()\n        (loss, (recon, metrics)) = tokenizer_loss_fn(tokenizer, inputs, training=False)\n        return loss, recon, metrics\n\n    def calculate_validation_metrics(val_dataloader, tokenizer):\n        step = 0\n        loss_per_step = []\n        metrics_per_step = []\n        batch = None\n        recon = None\n        for batch in val_dataloader:\n            loss, recon, metrics = val_step(tokenizer, batch)\n            loss_per_step.append(loss)\n            metrics_per_step.append(metrics)\n            step += 1\n            if step > args.val_steps:\n                break\n\n        if step < args.val_steps:\n            print(\n                f""Warning: Your validation dataset is too small to make val_steps many steps. Made {step} steps, expected {args.val_steps}""\n            )\n\n        val_loss = np.mean(loss_per_step)\n        val_metrics = {\n            f""val_{key}"": np.mean([float(m[key]) for m in metrics_per_step])\n            for key in metrics_per_step[0].keys()\n        }\n        val_metrics[""val_loss""] = val_loss\n        return val_metrics, batch, recon\n\n    # --- TRAIN LOOP ---\n    dataloader_train = (\n        {\n            ""videos"": jax.make_array_from_process_local_data(\n                videos_sharding, elem[""videos""]\n            ),\n        }\n        for elem in train_iterator\n    )\n    dataloader_val = None\n    if val_iterator:\n        dataloader_val = (\n            {\n                ""videos"": jax.make_array_from_process_local_data(\n                    videos_sharding, elem[""videos""]\n                ),\n            }\n            for elem in val_iterator\n        )\n    if jax.process_index() == 0:\n        first_batch = next(dataloader_train)\n        compiled = train_step.lower(optimizer, first_batch).compile()\n        print_compiled_memory_stats(compiled.memory_analysis())\n        print_compiled_cost_analysis(compiled.cost_analysis())\n        # Do not skip the first batch during training\n        dataloader_train = itertools.chain([first_batch], dataloader_train)\n    print(f""Starting training from step {step}..."")\n    first_step = step\n    while step < args.num_steps:\n        for batch in dataloader_train:\n            # --- Train step ---\n            loss, recon, metrics = train_step(optimizer, batch)\n            if step == first_step:\n                print_mem_stats(""After params initialized"")\n            step += 1\n\n            # --- Validation loss ---\n            val_results = {}\n            if dataloader_val and step % args.val_interval == 0:\n                print(""Calculating validation metrics..."")\n                val_metrics, val_gt_batch, val_recon = calculate_validation_metrics(\n                    dataloader_val, optimizer.model\n                )\n                print(f""Step {step}, validation loss: {val_metrics['val_loss']}"")\n                val_results = {\n                    ""metrics"": val_metrics,\n                    ""gt_batch"": val_gt_batch,\n                    ""recon"": val_recon,\n                }\n\n            # --- Logging ---\n            if args.log:\n                if step % args.log_interval == 0 and jax.process_index() == 0:\n                    log_dict = {""loss"": loss, ""step"": step, **metrics}\n                    if val_results:\n                        log_dict.update(val_results[""metrics""])\n                    wandb.log(log_dict)\n                if step % args.log_image_interval == 0:\n                    gt_seq = batch[""videos""][0].astype(jnp.float32) / 255.0\n                    recon_seq = recon[0].clip(0, 1)\n                    comparison_seq = jnp.concatenate((gt_seq, recon_seq), axis=1)\n                    comparison_seq = einops.rearrange(\n                        comparison_seq * 255, ""t h w c -> h (t w) c""\n                    )\n                    if val_results and step % args.val_interval == 0:\n                        val_results[""gt_seq_val""] = (\n                            val_results[""gt_batch""][""videos""][0].astype(jnp.float32)\n                            / 255.0\n                        )\n                        val_results[""recon_seq_val""] = val_results[""recon""][0].clip(\n                            0, 1\n                        )\n                        val_results[""val_comparison_seq""] = jnp.concatenate(\n                            (val_results[""gt_seq_val""], val_results[""recon_seq_val""]),\n                            axis=1,\n                        )\n                        val_results[""val_comparison_seq""] = einops.rearrange(\n                            val_results[""val_comparison_seq""] * 255,\n                            ""t h w c -> h (t w) c"",\n                        )\n                    # NOTE: Process-dependent control flow deliberately happens\n                    # after indexing operation since it must not contain code\n                    # sections that lead to cross-accelerator communication.\n                    if jax.process_index() == 0:\n                        log_images = dict(\n                            image=wandb.Image(np.asarray(gt_seq[0])),\n                            recon=wandb.Image(np.asarray(recon_seq[0])),\n                            true_vs_recon=wandb.Image(\n                                np.asarray(comparison_seq.astype(np.uint8))\n                            ),\n                        )\n                        if val_results and step % args.val_interval == 0:\n                            log_images.update(\n                                dict(\n                                    val_image=wandb.Image(\n                                        np.asarray(val_results[""gt_seq_val""][0])\n                                    ),\n                                    val_recon=wandb.Image(\n                                        np.asarray(val_results[""recon_seq_val""][0])\n                                    ),\n                                    val_true_vs_recon=wandb.Image(\n                                        np.asarray(\n                                            val_results[""val_comparison_seq""].astype(\n                                                np.uint8\n                                            )\n                                        )\n                                    ),\n                                )\n                            )\n                        wandb.log(log_images)\n            # --- Checkpointing ---\n            if args.save_ckpt and step % args.log_checkpoint_interval == 0:\n                assert checkpoint_manager is not None\n                optimizer_state = nnx.state(optimizer)\n                if val_iterator:\n                    ckpt_manager_args = ocp.args.Composite(\n                        model_state=ocp.args.PyTreeSave(optimizer_state),  # type: ignore\n                        train_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            train_iterator  # type: ignore\n                        ),\n                        val_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            val_iterator  # type: ignore\n                        ),\n                    )\n                else:\n                    ckpt_manager_args = ocp.args.Composite(\n                        model_state=ocp.args.PyTreeSave(optimizer_state),  # type: ignore\n                        train_dataloader_state=grain.checkpoint.CheckpointSave(  # type: ignore\n                            train_iterator  # type: ignore\n                        ),\n                    )\n                checkpoint_manager.save(step, args=ckpt_manager_args)\n                print(f""Saved checkpoint at step {step}"")\n            if step >= args.num_steps:\n                break\n\n    if checkpoint_manager:\n        checkpoint_manager.close()\n\n\nif __name__ == ""__main__"":\n    args = tyro.cli(Args)\n    main(args)\n",python,tab
+2,635,"jasmine/train_tokenizer.py",432,0,"",python,selection_mouse
+3,636,"jasmine/train_tokenizer.py",10,0,"",python,selection_command
+4,641,"jasmine/train_tokenizer.py",431,0,"",python,selection_command
+5,757,"jasmine/train_tokenizer.py",314,0,"",python,selection_mouse
+6,762,"jasmine/train_tokenizer.py",313,0,"",python,selection_command
+7,1705,"jasmine/train_tokenizer.py",345,0,"",python,selection_mouse
+8,1851,"jasmine/train_tokenizer.py",344,0,"",python,selection_command
+9,1873,"jasmine/train_tokenizer.py",385,0,"",python,selection_mouse
+10,1875,"jasmine/train_tokenizer.py",384,0,"",python,selection_command
+11,2600,"extension-output-pdoom-org.crowd-code-#1-crowd-code",0,0,"11:57:29 AM [info] Activating crowd-code\n11:57:29 AM [info] Recording started\n11:57:29 AM [info] Initializing git provider using file system watchers...\n11:57:30 AM [info] Git repository found\n11:57:30 AM [info] Git provider initialized successfully\n11:57:30 AM [info] Initial git state: [object Object]\n",Log,tab
+12,4062,"jasmine/train_tokenizer.py",0,0,"",python,tab
+13,4663,"jasmine/train_tokenizer.py",445,0,"",python,selection_mouse
+14,4767,"jasmine/train_tokenizer.py",444,0,"",python,selection_command
+15,14740,"TERMINAL",0,0,"undefined[tum_cte0515@hkn1990 jasmine]$ git branch",,terminal_command
+16,14803,"TERMINAL",0,0,"]633;C[?1h=\r  ablation/full-precision-training\r\n  ablation/use-pytorch-dataloader\r\n  action-mapper\r\n  add-noise-to-combat-exposure-bias\r\n  add-wandb-name-and-tags\r\n  before-nnx\r\n  causal-mem-reduce\r\n  causal-spatiotemporal-kv-cache\r\n:",,terminal_output
+17,15930,"TERMINAL",0,0,"\r[?1l>]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+18,17304,"TERMINAL",0,0,"branch",,terminal_command
+19,17310,"TERMINAL",0,0,"]633;Cprepend-action-maskgit\r\n]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+20,279470,"jasmine/train_tokenizer.py",0,0,"",python,tab
+21,432951,"TERMINAL",0,0,"git diff",,terminal_command
+22,433028,"TERMINAL",0,0,"]633;C[?1h=\r\r[?1l>]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+23,444780,"TERMINAL",0,0,"",,terminal_focus
+24,458977,"TERMINAL",0,0,"queue",,terminal_command
+25,459031,"TERMINAL",0,0,"]633;C",,terminal_output
+26,459129,"TERMINAL",0,0,"[?1049h(B[?7hEvery 1.0s: squeue --mehkn1990.localdomain: Sun Oct 19 12:05:08 2025JOBID PARTITION     NAME     USER ST\tTIME  NODES NODELIST(REASON)",,terminal_output
+27,459900,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+28,464279,"TERMINAL",0,0,"idling",,terminal_command
+29,464374,"TERMINAL",0,0,"]633;C[?1049h(B[?7hEvery 1.0s: sinfo_t_idlehkn1990.localdomain: Sun Oct 19 12:05:13 2025Partition dev_cpuonly:     11 nodes idle\rPartition cpuonly:    115 nodes idle\rPartition dev_accelerated:\t 3 nodes idle\rPartition accelerated:     65 nodes idle\rPartition dev_accelerated-h100    :\t 0 nodes idle\rPartition accelerated-h100:\t 0 nodes idle\rPartition large:\t 7 nodes idle\rPartition accelerated-h200:\t 5 nodes idle",,terminal_output
+30,465417,"TERMINAL",0,0,"4",,terminal_output
+31,466334,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+32,473648,"TERMINAL",0,0,"source .venv/bin/activate",,terminal_command
+33,473672,"TERMINAL",0,0,"]633;C]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+34,512244,"TERMINAL",0,0,"git checkout -b 'ablation/flash-attn-only-spatial'",,terminal_command
+35,512296,"TERMINAL",0,0,"]633;C",,terminal_output
+36,512484,"TERMINAL",0,0,"Switched to a new branch 'ablation/flash-attn-only-spatial'\r\n]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+37,515969,"",0,0,"Switched from branch 'prepend-action-maskgit' to 'ablation/flash-attn-only-spatial'",,git_branch_checkout
+38,516511,"jasmine/utils/nn.py",0,0,"import math\nfrom typing import Tuple, Callable, List\n\nfrom flax import nnx\nimport jax\nimport jax.numpy as jnp\nimport einops\n\n\ndef _get_spatiotemporal_positional_encoding(d_model: int, max_len: int = 5000):\n    """"""\n    Creates a function that applies separate sinusoidal positional encodings to the temporal and spatial dimensions.\n    """"""\n    pe = jnp.zeros((max_len, d_model))\n    position = jnp.arange(0, max_len, dtype=jnp.float32)[:, None]\n    div_term = jnp.exp(jnp.arange(0, d_model, 2) * (-math.log(10000.0) / d_model))\n    pe = pe.at[:, 0::2].set(jnp.sin(position * div_term))\n    pe = pe.at[:, 1::2].set(jnp.cos(position * div_term))\n\n    def _encode(x: jax.Array) -> jax.Array:\n        """"""\n        Args:\n            x: The input tensor of shape (Batch, Time, Space, Dimension).\n\n        Returns:\n            The input tensor with positional encodings added.\n        """"""\n        assert x.ndim == 4, f""Input must be 4-dimensional, but got shape {x.shape}""\n\n        num_timesteps = x.shape[1]\n        num_spatial_patches = x.shape[2]\n\n        # Temporal positional encoding: (1, T, 1, D)\n        temporal_pe = pe[None, :num_timesteps, None, :]\n        x = x + temporal_pe\n\n        # Spatial positional encoding: (1, 1, S, D)\n        spatial_pe = pe[None, None, :num_spatial_patches, :]\n        x = x + spatial_pe\n\n        return x\n\n    return _encode\n\n\nclass STBlock(nnx.Module):\n    def __init__(\n        self,\n        dim: int,\n        ffn_dim: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        rngs: nnx.Rngs,\n        sow_weights: bool,\n        sow_activations: bool,\n    ):\n        self.dim = dim\n        self.ffn_dim = ffn_dim\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.spatial_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.spatial_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.dim,\n            qkv_features=self.dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=False\n            ),\n            rngs=rngs,\n            decode=False,\n        )\n\n        self.temporal_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.temporal_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.dim,\n            qkv_features=self.dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=True\n            ),\n            rngs=rngs,\n            decode=False,\n        )\n\n        self.ffn_norm = nnx.LayerNorm(\n            num_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.ffn_dense1 = nnx.Linear(\n            in_features=self.dim,\n            out_features=self.ffn_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_dense2 = nnx.Linear(\n            in_features=self.ffn_dim,\n            out_features=self.dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    @nnx.remat\n    def __call__(self, x_BTNM: jax.Array) -> jax.Array:\n        # --- Spatial attention ---\n        z_BTNM = self.spatial_norm(x_BTNM)\n        z_BTNM = self.spatial_attention(z_BTNM, sow_weights=self.sow_weights)\n        x_BTNM = x_BTNM + z_BTNM\n\n        # --- Temporal attention ---\n        x_BNTM = x_BTNM.swapaxes(1, 2)\n        z_BNTM = self.temporal_norm(x_BNTM)\n        z_BNTM = self.temporal_attention(z_BNTM, sow_weights=self.sow_weights)\n        x_BNTM = x_BNTM + z_BNTM\n        x_BTNM = x_BNTM.swapaxes(1, 2)\n\n        # --- Feedforward ---\n        z_BTNM = self.ffn_norm(x_BTNM)\n        z_BTND = self.ffn_dense1(z_BTNM)\n        z_BTND = jax.nn.gelu(z_BTND)\n        z_BTNM = self.ffn_dense2(z_BTND)\n        x_BTNM = x_BTNM + z_BTNM\n        if self.sow_activations:\n            self.sow(nnx.Intermediate, ""activations"", x_BTNM)\n        return x_BTNM\n\n\nclass STTransformer(nnx.Module):\n    """"""\n    Dimension keys:\n        B: batch size\n        T: number of frames\n        N: number of patches per frame\n        I: number of input features\n        M: model dimension\n        D: FFN dimension\n        V: vocabulary size\n    """"""\n\n    def __init__(\n        self,\n        input_dim: int,\n        model_dim: int,\n        ffn_dim: int,\n        out_dim: int,\n        num_blocks: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        rngs: nnx.Rngs,\n        sow_weights: bool = False,\n        sow_activations: bool = False,\n        sow_logits: bool = False,\n        max_len: int = 5000,\n    ):\n        self.input_dim = input_dim\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.out_dim = out_dim\n        self.num_blocks = num_blocks\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.sow_logits = sow_logits\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.input_norm1 = nnx.LayerNorm(\n            num_features=self.input_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.input_dense = nnx.Linear(\n            in_features=self.input_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_norm2 = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n\n        self.pos_enc = _get_spatiotemporal_positional_encoding(\n            self.model_dim, max_len=max_len\n        )\n\n        self.blocks = []\n        for _ in range(self.num_blocks):\n            self.blocks.append(\n                STBlock(\n                    dim=self.model_dim,\n                    ffn_dim=self.ffn_dim,\n                    num_heads=self.num_heads,\n                    dropout=self.dropout,\n                    param_dtype=self.param_dtype,\n                    dtype=self.dtype,\n                    use_flash_attention=self.use_flash_attention,\n                    rngs=rngs,\n                    sow_weights=self.sow_weights,\n                    sow_activations=self.sow_activations,\n                )\n            )\n\n        self.output_dense = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.out_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    def __call__(self, x_BTNI: jax.Array) -> jax.Array:\n        x_BTNI = self.input_norm1(x_BTNI)\n        x_BTNM = self.input_dense(x_BTNI)\n        x_BTNM = self.input_norm2(x_BTNM)\n        x_BTNM = self.pos_enc(x_BTNM)\n        for block in self.blocks:\n            x_BTNM = block(x_BTNM)\n\n        x_BTNV = self.output_dense(x_BTNM)\n        if self.sow_logits:\n            self.sow(nnx.Intermediate, ""logits"", x_BTNV)\n        return x_BTNV\n\n\nclass TransformerBlock(nnx.Module):\n    def __init__(\n        self,\n        model_dim: int,\n        ffn_dim: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        decode: bool,\n        rngs: nnx.Rngs,\n        sow_weights: bool,\n        sow_activations: bool,\n    ):\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.decode = decode\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.temporal_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.spatial_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.ffn_norm = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.temporal_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.model_dim,\n            qkv_features=self.model_dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=True\n            ),\n            rngs=rngs,\n            decode=self.decode,\n        )\n        self.spatial_attention = nnx.MultiHeadAttention(\n            num_heads=self.num_heads,\n            in_features=self.model_dim,\n            qkv_features=self.model_dim,\n            dropout_rate=self.dropout,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            attention_fn=_create_flash_attention_fn(\n                self.use_flash_attention, is_causal=True\n            ),\n            rngs=rngs,\n            decode=self.decode,\n        )\n        self.ffn_dense1 = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.ffn_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.ffn_dense2 = nnx.Linear(\n            in_features=self.ffn_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    @nnx.remat\n    def __call__(\n        self, x_BTNM: jax.Array, pos_index: Tuple[jax.Array, jax.Array] | None = None\n    ) -> jax.Array:\n        # --- Spatial attention ---\n        B, T, N, M = x_BTNM.shape\n        z_FNM = einops.rearrange(x_BTNM, ""b t n m -> (b t) n m"")\n        z_FNM = self.spatial_norm(z_FNM)\n        z_FNM = self.spatial_attention(z_FNM, sow_weights=self.sow_weights)\n        z_BTNM = einops.rearrange(z_FNM, ""(b t) n m -> b t n m"", t=T)\n        x_BTNM = x_BTNM + z_BTNM\n        # --- Temporal attention ---\n        z_PTM = einops.rearrange(x_BTNM, ""b t n m -> (b n) t m"")\n        z_PTM = self.temporal_norm(z_PTM)\n        z_PTM = self.temporal_attention(z_PTM, sow_weights=self.sow_weights)\n        z_BTNM = einops.rearrange(z_PTM, ""(b n) t m -> b t n m"", n=N)\n        x_BTNM = x_BTNM + z_BTNM\n        # --- Feedforward ---\n        z_BTNM = self.ffn_norm(x_BTNM)\n        z_BTND = self.ffn_dense1(z_BTNM)\n        z_BTND = jax.nn.gelu(z_BTND)\n        z_BTNM = self.ffn_dense2(z_BTND)\n        x_BTNM = x_BTNM + z_BTNM\n        if self.sow_activations:\n            self.sow(nnx.Intermediate, ""activations"", x_BTNM)\n\n        return x_BTNM\n\n\nclass Transformer(nnx.Module):\n    """"""\n    Dimension keys:\n        B: batch size\n        T: number of frames\n        N: number of patches per frame\n        I: number of input features\n        M: model dimension\n        D: FFN dimension\n        V: vocabulary size\n        F: number of frames in batch\n        P: number of patch positions in batch\n    """"""\n\n    def __init__(\n        self,\n        input_dim: int,\n        model_dim: int,\n        ffn_dim: int,\n        out_dim: int,\n        num_blocks: int,\n        num_heads: int,\n        dropout: float,\n        param_dtype: jnp.dtype,\n        dtype: jnp.dtype,\n        use_flash_attention: bool,\n        decode: bool,\n        rngs: nnx.Rngs,\n        sow_logits: bool = False,\n        sow_weights: bool = False,\n        sow_activations: bool = False,\n        max_len: int = 5000,\n    ):\n        self.input_dim = input_dim\n        self.model_dim = model_dim\n        self.ffn_dim = ffn_dim\n        self.out_dim = out_dim\n        self.num_blocks = num_blocks\n        self.num_heads = num_heads\n        self.dropout = dropout\n        self.param_dtype = param_dtype\n        self.dtype = dtype\n        self.use_flash_attention = use_flash_attention\n        self.sow_logits = sow_logits\n        self.sow_weights = sow_weights\n        self.sow_activations = sow_activations\n\n        self.input_norm1 = nnx.LayerNorm(\n            num_features=self.input_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n        self.input_dense = nnx.Linear(\n            in_features=self.input_dim,\n            out_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n        self.input_norm2 = nnx.LayerNorm(\n            num_features=self.model_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.param_dtype,  # layer norm in full precision\n            rngs=rngs,\n        )\n\n        self.pos_enc = _get_spatiotemporal_positional_encoding(\n            self.model_dim, max_len=max_len\n        )\n\n        self.blocks: List[TransformerBlock] = []\n        for _ in range(self.num_blocks):\n            self.blocks.append(\n                TransformerBlock(\n                    model_dim=self.model_dim,\n                    ffn_dim=self.ffn_dim,\n                    num_heads=self.num_heads,\n                    dropout=self.dropout,\n                    param_dtype=self.param_dtype,\n                    dtype=self.dtype,\n                    use_flash_attention=self.use_flash_attention,\n                    decode=decode,\n                    sow_weights=self.sow_weights,\n                    sow_activations=self.sow_activations,\n                    rngs=rngs,\n                )\n            )\n        self.output_dense = nnx.Linear(\n            in_features=self.model_dim,\n            out_features=self.out_dim,\n            param_dtype=self.param_dtype,\n            dtype=self.dtype,\n            rngs=rngs,\n        )\n\n    def __call__(\n        self, x_BTNI: jax.Array, pos_index: Tuple[jax.Array, jax.Array] | None = None\n    ) -> jax.Array:\n        x_BTNI = self.input_norm1(x_BTNI)\n        x_BTNM = self.input_dense(x_BTNI)\n        x_BTNM = self.input_norm2(x_BTNM)\n        x_BTNM = self.pos_enc(x_BTNM)\n        for block in self.blocks:\n            x_BTNM = block(x_BTNM, pos_index)\n\n        x_BTNV = self.output_dense(x_BTNM)\n        if self.sow_logits:\n            self.sow(nnx.Intermediate, ""logits"", x_BTNV)\n        return x_BTNV\n\n\ndef normalize(x: jax.Array) -> jax.Array:\n    return x / (jnp.linalg.norm(x, ord=2, axis=-1, keepdims=True) + 1e-8)\n\n\nclass VectorQuantizer(nnx.Module):\n    """"""\n    Dimension keys:\n        D: B * T * N\n        K: number of latents\n        L: latent dimension\n    """"""\n\n    def __init__(\n        self,\n        latent_dim: int,\n        num_latents: int,\n        dropout: float,\n        dtype: jnp.dtype,\n        rngs: nnx.Rngs,\n    ):\n        self.latent_dim = latent_dim\n        self.num_latents = num_latents\n        self.dropout = dropout\n        self.dtype = dtype\n\n        self.codebook = nnx.Param(\n            normalize(\n                nnx.initializers.normal(stddev=1)(\n                    rngs.params(), (self.num_latents, self.latent_dim)\n                )\n            )\n        )\n        self.drop = nnx.Dropout(self.dropout, rngs=rngs)\n\n    def __call__(\n        self, x_DL: jax.Array, training: bool\n    ) -> Tuple[jax.Array, jax.Array, jax.Array, jax.Array]:\n        # --- Compute distances ---\n        x_DL = x_DL.astype(self.dtype)\n        codebook = self.codebook.value.astype(self.dtype)\n\n        x_DL = normalize(x_DL)\n        normalized_codebook_KL = normalize(codebook)\n        distance_DK = -jnp.matmul(x_DL, normalized_codebook_KL.T)\n        if training:\n            distance_DK = self.drop(distance_DK)\n\n        # --- Get indices and embeddings ---\n        indices_D = jnp.argmin(distance_DK, axis=-1)\n        z_DL = codebook[indices_D]\n\n        # --- Straight through estimator ---\n        z_q_DL = x_DL + jax.lax.stop_gradient(z_DL - x_DL)\n        return z_q_DL, z_DL, x_DL, indices_D\n\n    def get_codes(self, indices_E: jax.Array) -> jax.Array:\n        return self.codebook[indices_E]\n\n\ndef _create_flash_attention_fn(use_flash_attention: bool, is_causal: bool) -> Callable:\n    """"""\n    Create an attention function that uses flash attention if enabled.\n\n    flax.nnx.MultiHeadAttention provides tensors with shape (batch..., length, num_heads, head_dim),\n    but jax.nn.dot_product_attention expects (batch, length, num_heads, head_dim). We reshape to\n    ensure compatibility. cuDNN's flash attention additionally requires a sequence length that\n    is a multiple of 4. We pad the sequence length to the nearest multiple of 4 and mask\n    accordingly. Note that cuDNN requires the mask to be broadcast before calling the attention\n    function due to strict shape checking.\n    """"""\n\n    def attention_fn(\n        query_BTHD, key_BSHD, value_BSHD, bias=None, mask_B111=None, **kwargs\n    ):\n        implementation = ""cudnn"" if use_flash_attention else None\n\n        def _merge_batch_dims(x):\n            return einops.rearrange(x, ""... l h k -> (...) l h k"")\n\n        def _pad(x, pad_size):\n            return jnp.pad(x, ((0, 0), (0, pad_size), (0, 0), (0, 0)))\n\n        original_shape = query_BTHD.shape\n        T = query_BTHD.shape[-3]\n        S = key_BSHD.shape[-3]\n\n        # Pad to nearest multiple of 4\n        Q = ((T + 3) // 4) * 4\n        pad_size_Q = Q - T\n        K = ((S + 3) // 4) * 4\n        pad_size_K = K - S\n\n        query_BQHD = _pad(_merge_batch_dims(query_BTHD), pad_size_Q)\n        key_BKHD = _pad(_merge_batch_dims(key_BSHD), pad_size_K)\n        value_BKHD = _pad(_merge_batch_dims(value_BSHD), pad_size_K)\n\n        attention_mask = jnp.ones((Q, K), dtype=jnp.bool_)\n        attention_mask = attention_mask.at[T:, :].set(False)\n        attention_mask = attention_mask.at[:, S:].set(False)\n\n        mask_11TS = attention_mask[jnp.newaxis, jnp.newaxis, :, :]\n\n        bias_4d = (\n            jnp.pad(\n                _merge_batch_dims(bias),\n                ((0, 0), (0, 0), (0, pad_size_Q), (0, pad_size_K)),\n            )\n            if bias is not None\n            else None\n        )\n\n        # NOTE: jax.nn.dot_product_attention does not support dropout\n        output_4d = jax.nn.dot_product_attention(\n            query=query_BQHD,\n            key=key_BKHD,\n            value=value_BKHD,\n            bias=bias_4d,\n            mask=mask_11TS,\n            implementation=implementation,\n            is_causal=is_causal,\n        )\n        return output_4d[..., :T, :, :].reshape(original_shape)\n\n    return attention_fn\n",python,tab
+39,517069,"jasmine/utils/nn.py",12,0,"",python,selection_command
+40,518308,"jasmine/utils/nn.py",53,0,"",python,selection_command
+41,518411,"jasmine/utils/nn.py",54,0,"",python,selection_command
+42,518560,"jasmine/utils/nn.py",75,0,"",python,selection_command
+43,518713,"jasmine/utils/nn.py",86,0,"",python,selection_command
+44,518875,"jasmine/utils/nn.py",110,0,"",python,selection_command
+45,518975,"jasmine/utils/nn.py",124,0,"",python,selection_command
+46,519929,"jasmine/utils/nn.py",125,0,"",python,selection_command
+47,521426,"jasmine/utils/nn.py",126,0,"",python,selection_command
+48,521540,"jasmine/utils/nn.py",206,0,"",python,selection_command
+49,521662,"jasmine/utils/nn.py",214,0,"",python,selection_command
+50,521821,"jasmine/utils/nn.py",331,0,"",python,selection_command
+51,521926,"jasmine/utils/nn.py",339,0,"",python,selection_command
+52,522057,"jasmine/utils/nn.py",378,0,"",python,selection_command
+53,522193,"jasmine/utils/nn.py",444,0,"",python,selection_command
+54,522305,"jasmine/utils/nn.py",527,0,"",python,selection_command
+55,522488,"jasmine/utils/nn.py",585,0,"",python,selection_command
+56,522700,"jasmine/utils/nn.py",643,0,"",python,selection_command
+57,522813,"jasmine/utils/nn.py",644,0,"",python,selection_command
+58,522959,"jasmine/utils/nn.py",688,0,"",python,selection_command
+59,523116,"jasmine/utils/nn.py",700,0,"",python,selection_command
+60,523259,"jasmine/utils/nn.py",714,0,"",python,selection_command
+61,523410,"jasmine/utils/nn.py",788,0,"",python,selection_command
+62,523583,"jasmine/utils/nn.py",789,0,"",python,selection_command
+63,523720,"jasmine/utils/nn.py",806,0,"",python,selection_command
+64,523865,"jasmine/utils/nn.py",868,0,"",python,selection_command
+65,524022,"jasmine/utils/nn.py",880,0,"",python,selection_command
+66,524167,"jasmine/utils/nn.py",964,0,"",python,selection_command
+67,525298,"jasmine/utils/nn.py",1151,0,"",python,selection_command
+68,525874,"jasmine/utils/nn.py",1321,0,"",python,selection_command
+69,526417,"jasmine/utils/nn.py",1387,0,"",python,selection_command
+70,526853,"jasmine/utils/nn.py",1507,0,"",python,selection_command
+71,527153,"jasmine/utils/nn.py",1682,0,"",python,selection_command
+72,527469,"jasmine/utils/nn.py",1848,0,"",python,selection_command
+73,527747,"jasmine/utils/nn.py",2060,0,"",python,selection_command
+74,528035,"jasmine/utils/nn.py",2295,0,"",python,selection_command
+75,528339,"jasmine/utils/nn.py",2513,0,"",python,selection_command
+76,529378,"jasmine/utils/nn.py",2698,0,"",python,selection_command
+77,531209,"jasmine/utils/nn.py",2911,0,"",python,selection_command
+78,531882,"jasmine/utils/nn.py",3125,0,"",python,selection_command
+79,535244,"jasmine/utils/nn.py",3223,0,"",python,selection_mouse
+80,536448,"jasmine/utils/nn.py",3282,0,"",python,selection_mouse
+81,537218,"jasmine/utils/nn.py",3266,0,"",python,selection_mouse
+82,537481,"jasmine/utils/nn.py",3266,2,"se",python,selection_mouse
+83,537482,"jasmine/utils/nn.py",3266,10,"self.use_f",python,selection_mouse
+84,537660,"jasmine/utils/nn.py",3276,0,"",python,selection_mouse
+85,539043,"jasmine/utils/nn.py",3231,0,"",python,selection_mouse
+86,539088,"jasmine/utils/nn.py",17439,0,"",python,selection_command
+87,540327,"jasmine/utils/nn.py",17530,0,"",python,selection_mouse
+88,540335,"jasmine/utils/nn.py",17529,0,"",python,selection_command
+89,542608,"jasmine/sample.py",0,0,"from dataclasses import dataclass\nimport time\nimport os\nimport optax\n\nimport dm_pix as pix\nimport einops\nimport jax\nimport jax.numpy as jnp\nimport flax.linen as nn\nimport numpy as np\nimport orbax.checkpoint as ocp\nfrom PIL import Image, ImageDraw\nimport tyro\nfrom flax import nnx\n\nfrom genie import Genie\nfrom utils.dataloader import get_dataloader\n\n\n@dataclass\nclass Args:\n    # Experiment\n    seed: int = 0\n    seq_len: int = 16\n    image_channels: int = 3\n    image_height: int = 90\n    image_width: int = 160\n    data_dir: str = ""data/coinrun_episodes""\n    checkpoint: str = """"\n    print_action_indices: bool = True\n    output_dir: str = ""gifs/""\n    # Sampling\n    batch_size: int = 1\n    maskgit_steps: int = 25\n    temperature: float = 1.0\n    sample_argmax: bool = True\n    start_frame: int = 1\n    # Tokenizer checkpoint\n    tokenizer_dim: int = 512\n    tokenizer_ffn_dim: int = 2048\n    latent_patch_dim: int = 32\n    num_patch_latents: int = 1024\n    patch_size: int = 4\n    tokenizer_num_blocks: int = 4\n    tokenizer_num_heads: int = 8\n    # LAM checkpoint\n    lam_dim: int = 512\n    lam_ffn_dim: int = 2048\n    latent_action_dim: int = 32\n    num_actions: int = 6\n    lam_patch_size: int = 16\n    lam_num_blocks: int = 4\n    lam_num_heads: int = 8\n    use_gt_actions: bool = False\n    # Dynamics checkpoint\n    dyna_type: str = ""maskgit""\n    dyna_dim: int = 512\n    dyna_ffn_dim: int = 2048\n    dyna_num_blocks: int = 6\n    dyna_num_heads: int = 8\n    param_dtype = jnp.float32\n    dtype = jnp.bfloat16\n    use_flash_attention: bool = True\n\n\nargs = tyro.cli(Args)\n\nif __name__ == ""__main__"":\n    """"""\n    Dimension keys:\n        B: batch size\n        T: number of input (conditioning) frames\n        N: number of patches per frame\n        S: sequence length\n        H: height\n        W: width\n        E: B * (S - 1)\n    """"""\n    jax.distributed.initialize()\n\n    rng = jax.random.key(args.seed)\n\n    # --- Load Genie checkpoint ---\n    rngs = nnx.Rngs(rng)\n    genie = Genie(\n        # Tokenizer\n        in_dim=args.image_channels,\n        tokenizer_dim=args.tokenizer_dim,\n        tokenizer_ffn_dim=args.tokenizer_ffn_dim,\n        latent_patch_dim=args.latent_patch_dim,\n        num_patch_latents=args.num_patch_latents,\n        patch_size=args.patch_size,\n        tokenizer_num_blocks=args.tokenizer_num_blocks,\n        tokenizer_num_heads=args.tokenizer_num_heads,\n        # LAM\n        lam_dim=args.lam_dim,\n        lam_ffn_dim=args.lam_ffn_dim,\n        latent_action_dim=args.latent_action_dim,\n        num_actions=args.num_actions,\n        lam_patch_size=args.lam_patch_size,\n        lam_num_blocks=args.lam_num_blocks,\n        lam_num_heads=args.lam_num_heads,\n        lam_co_train=False,\n        use_gt_actions=args.use_gt_actions,\n        # Dynamics\n        dyna_type=args.dyna_type,\n        dyna_dim=args.dyna_dim,\n        dyna_ffn_dim=args.dyna_ffn_dim,\n        dyna_num_blocks=args.dyna_num_blocks,\n        dyna_num_heads=args.dyna_num_heads,\n        param_dtype=args.param_dtype,\n        dtype=args.dtype,\n        use_flash_attention=args.use_flash_attention,\n        # FIXME (f.srambical): implement spatiotemporal KV caching and set decode=True\n        decode=False,\n        rngs=rngs,\n    )\n\n    # Need to delete lam decoder for checkpoint loading\n    if not args.use_gt_actions:\n        assert genie.lam is not None\n        del genie.lam.decoder\n\n    handler_registry = ocp.handlers.DefaultCheckpointHandlerRegistry()\n    handler_registry.add(\n        ""model_state"", ocp.args.PyTreeSave, ocp.handlers.PyTreeCheckpointHandler\n    )\n    handler_registry.add(\n        ""model_state"", ocp.args.PyTreeRestore, ocp.handlers.PyTreeCheckpointHandler\n    )\n    checkpoint_options = ocp.CheckpointManagerOptions(\n        step_format_fixed_length=6,\n    )\n    checkpoint_manager = ocp.CheckpointManager(\n        args.checkpoint,\n        options=checkpoint_options,\n        handler_registry=handler_registry,\n    )\n\n    dummy_tx = optax.adamw(\n        learning_rate=optax.linear_schedule(0.0001, 0.0001, 10000),\n        b1=0.9,\n        b2=0.9,\n        weight_decay=1e-4,\n        mu_dtype=args.dtype,\n    )\n    dummy_optimizer = nnx.ModelAndOptimizer(genie, dummy_tx)\n\n    abstract_optimizer = nnx.eval_shape(lambda: dummy_optimizer)\n    abstract_optimizer_state = nnx.state(abstract_optimizer)\n    restored = checkpoint_manager.restore(\n        checkpoint_manager.latest_step(),\n        args=ocp.args.Composite(\n            model_state=ocp.args.PyTreeRestore(abstract_optimizer_state),  # type: ignore\n        ),\n    )\n    restored_optimizer_state = restored[""model_state""]\n    nnx.update(dummy_optimizer, restored_optimizer_state)\n\n    # --- Define sampling function ---\n    def _sampling_fn(model: Genie, batch: dict) -> jax.Array:\n        """"""Runs Genie.sample with pre-defined generation hyper-parameters.""""""\n        assert args.dyna_type in [\n            ""maskgit"",\n            ""causal"",\n        ], f""Invalid dynamics type: {args.dyna_type}""\n        frames, _ = model.sample(\n            batch,\n            args.seq_len,\n            args.temperature,\n            args.sample_argmax,\n            args.maskgit_steps,\n        )\n        return frames\n\n    # --- Define autoregressive sampling loop ---\n    def _autoreg_sample(genie, rng, batch):\n        batch[""videos""] = batch[""videos""][:, : args.start_frame]\n        batch[""rng""] = rng\n        generated_vid_BSHWC = _sampling_fn(genie, batch)\n        return generated_vid_BSHWC\n\n    # --- Get video + latent actions ---\n    array_record_files = [\n        os.path.join(args.data_dir, x)\n        for x in os.listdir(args.data_dir)\n        if x.endswith("".array_record"")\n    ]\n    dataloader = get_dataloader(\n        array_record_files,\n        args.seq_len,\n        args.batch_size,\n        args.image_height,\n        args.image_width,\n        args.image_channels,\n        # We don't use workers in order to avoid grain shutdown issues (https://github.com/google/grain/issues/398)\n        num_workers=0,\n        prefetch_buffer_size=1,\n        seed=args.seed,\n    )\n    dataloader = iter(dataloader)\n    batch = next(dataloader)\n    gt_video = jnp.asarray(batch[""videos""], dtype=jnp.float32) / 255.0\n    batch[""videos""] = gt_video.astype(args.dtype)\n    # Get latent actions for all videos in the batch\n    action_batch_E = None\n    if not args.use_gt_actions:\n        action_batch_E = genie.vq_encode(batch, training=False)\n        batch[""latent_actions""] = action_batch_E\n\n    # --- Sample + evaluate video ---\n    recon_video_BSHWC = _autoreg_sample(genie, rng, batch)\n    recon_video_BSHWC = recon_video_BSHWC.astype(jnp.float32)\n\n    gt = gt_video.clip(0, 1)[:, args.start_frame :]\n    recon = recon_video_BSHWC.clip(0, 1)[:, args.start_frame :]\n\n    ssim_vmap = jax.vmap(pix.ssim, in_axes=(0, 0))\n    psnr_vmap = jax.vmap(pix.psnr, in_axes=(0, 0))\n    ssim = ssim_vmap(gt, recon)\n    psnr = psnr_vmap(gt, recon)\n    per_frame_ssim = ssim.mean(0)\n    per_frame_psnr = psnr.mean(0)\n    avg_ssim = ssim.mean()\n    avg_psnr = psnr.mean()\n\n    print(""Per-frame SSIM:\n"", per_frame_ssim)\n    print(""Per-frame PSNR:\n"", per_frame_psnr)\n\n    print(f""SSIM: {avg_ssim}"")\n    print(f""PSNR: {avg_psnr}"")\n\n    # --- Construct video ---\n    true_videos = (gt_video * 255).astype(np.uint8)\n    pred_videos = (recon_video_BSHWC * 255).astype(np.uint8)\n    video_comparison = np.zeros((2, *recon_video_BSHWC.shape), dtype=np.uint8)\n    video_comparison[0] = true_videos[:, : args.seq_len]\n    video_comparison[1] = pred_videos\n    frames = einops.rearrange(video_comparison, ""n b t h w c -> t (b h) (n w) c"")\n\n    # --- Save video ---\n    imgs = [Image.fromarray(img) for img in frames]\n    # Write actions on each frame, on each row (i.e., for each video in the batch, on the GT row)\n    B = batch[""videos""].shape[0]\n    if action_batch_E is not None:\n        action_batch_BSm11 = jnp.reshape(action_batch_E, (B, args.seq_len - 1, 1))\n    else:\n        action_batch_BSm11 = jnp.reshape(\n            batch[""actions""][:, :-1], (B, args.seq_len - 1, 1)\n        )\n    for t, img in enumerate(imgs[1:]):\n        d = ImageDraw.Draw(img)\n        for row in range(B):\n            if args.print_action_indices:\n                action = action_batch_BSm11[row, t, 0]\n                y_offset = row * batch[""videos""].shape[2] + 2\n                d.text((2, y_offset), f""{action}"", fill=255)\n\n    os.makedirs(args.output_dir, exist_ok=True)\n    imgs[0].save(\n        os.path.join(args.output_dir, f""generation_{time.time()}.gif""),\n        save_all=True,\n        append_images=imgs[1:],\n        duration=250,\n        loop=0,\n    )\n",python,tab
+90,543597,"jasmine/utils/nn.py",0,0,"",python,tab
+91,551731,"jasmine/utils/nn.py",10448,0,"",python,selection_mouse
+92,551909,"jasmine/utils/nn.py",10448,2,"se",python,selection_mouse
+93,551910,"jasmine/utils/nn.py",10448,4,"self",python,selection_mouse
+94,551939,"jasmine/utils/nn.py",10448,5,"self.",python,selection_mouse
+95,551940,"jasmine/utils/nn.py",10448,7,"self.us",python,selection_mouse
+96,551953,"jasmine/utils/nn.py",10448,10,"self.use_f",python,selection_mouse
+97,551967,"jasmine/utils/nn.py",10448,11,"self.use_fl",python,selection_mouse
+98,551989,"jasmine/utils/nn.py",10448,13,"self.use_flas",python,selection_mouse
+99,552002,"jasmine/utils/nn.py",10448,14,"self.use_flash",python,selection_mouse
+100,552020,"jasmine/utils/nn.py",10448,16,"self.use_flash_a",python,selection_mouse
+101,552068,"jasmine/utils/nn.py",10448,17,"self.use_flash_at",python,selection_mouse
+102,552069,"jasmine/utils/nn.py",10448,18,"self.use_flash_att",python,selection_mouse
+103,552085,"jasmine/utils/nn.py",10448,19,"self.use_flash_atte",python,selection_mouse
+104,552101,"jasmine/utils/nn.py",10448,20,"self.use_flash_atten",python,selection_mouse
+105,552178,"jasmine/utils/nn.py",10448,55,"self.use_flash_attention, is_causal=True\n            ),",python,selection_mouse
+106,552631,"jasmine/utils/nn.py",10448,25,"self.use_flash_attention,",python,selection_mouse
+107,552868,"jasmine/utils/nn.py",10448,24,"self.use_flash_attention",python,selection_mouse
+108,553782,"jasmine/utils/nn.py",10448,24,"",python,content
+109,554472,"jasmine/utils/nn.py",10448,0,"u",python,content
+110,554474,"jasmine/utils/nn.py",10449,0,"",python,selection_keyboard
+111,554540,"jasmine/utils/nn.py",10449,0,"s",python,content
+112,554541,"jasmine/utils/nn.py",10450,0,"",python,selection_keyboard
+113,554693,"jasmine/utils/nn.py",10450,0,"e",python,content
+114,554695,"jasmine/utils/nn.py",10451,0,"",python,selection_keyboard
+115,556942,"jasmine/utils/nn.py",10451,0,"_flash_attention",python,content
+116,558714,"jasmine/utils/nn.py",10467,0,"=",python,content
+117,558716,"jasmine/utils/nn.py",10468,0,"",python,selection_keyboard
+118,560273,"jasmine/utils/nn.py",10468,0,"F",python,content
+119,560274,"jasmine/utils/nn.py",10469,0,"",python,selection_keyboard
+120,560377,"jasmine/utils/nn.py",10469,0,"a",python,content
+121,560378,"jasmine/utils/nn.py",10470,0,"",python,selection_keyboard
+122,560605,"jasmine/utils/nn.py",10470,0,"l",python,content
+123,560607,"jasmine/utils/nn.py",10471,0,"",python,selection_keyboard
+124,561576,"jasmine/utils/nn.py",10471,0,"se",python,content
+125,563747,"jasmine/utils/nn.py",10472,0,"",python,selection_command
+126,566969,"jasmine/utils/nn.py",10449,0,"",python,selection_mouse
+127,567078,"jasmine/utils/nn.py",10449,2,"se",python,selection_mouse
+128,567099,"jasmine/utils/nn.py",10449,6,"se_fla",python,selection_mouse
+129,567123,"jasmine/utils/nn.py",10449,8,"se_flash",python,selection_mouse
+130,567142,"jasmine/utils/nn.py",10449,11,"se_flash_at",python,selection_mouse
+131,567154,"jasmine/utils/nn.py",10449,14,"se_flash_atten",python,selection_mouse
+132,567171,"jasmine/utils/nn.py",10449,16,"se_flash_attenti",python,selection_mouse
+133,567238,"jasmine/utils/nn.py",10449,18,"se_flash_attention",python,selection_mouse
+134,567239,"jasmine/utils/nn.py",10449,20,"se_flash_attention=F",python,selection_mouse
+135,567239,"jasmine/utils/nn.py",10449,21,"se_flash_attention=Fa",python,selection_mouse
+136,567240,"jasmine/utils/nn.py",10449,22,"se_flash_attention=Fal",python,selection_mouse
+137,567299,"jasmine/utils/nn.py",10449,23,"se_flash_attention=Fals",python,selection_mouse
+138,567334,"jasmine/utils/nn.py",10449,24,"se_flash_attention=False",python,selection_mouse
+139,567668,"jasmine/utils/nn.py",10449,23,"se_flash_attention=Fals",python,selection_mouse
+140,568495,"jasmine/utils/nn.py",10449,23,"",python,content
+141,568996,"jasmine/utils/nn.py",10449,0,"se_flash_attention=Fals",python,content
+142,569005,"jasmine/utils/nn.py",10449,0,"",python,selection_command
+143,569558,"jasmine/utils/nn.py",10448,25,"",python,content
+144,570022,"jasmine/utils/nn.py",10448,0,"self.use_flash_attention",python,content
+145,570028,"jasmine/utils/nn.py",10448,0,"",python,selection_command
+146,571427,"jasmine/utils/nn.py",10526,0,"",python,selection_mouse
+147,571457,"jasmine/utils/nn.py",10525,0,"",python,selection_command
+148,581715,"jasmine/utils/nn.py",9971,0,"",python,selection_mouse
+149,581775,"jasmine/utils/nn.py",9971,3,"sel",python,selection_mouse
+150,581799,"jasmine/utils/nn.py",9971,7,"self.us",python,selection_mouse
+151,581821,"jasmine/utils/nn.py",9971,8,"self.use",python,selection_mouse
+152,581836,"jasmine/utils/nn.py",9971,10,"self.use_f",python,selection_mouse
+153,581869,"jasmine/utils/nn.py",9971,12,"self.use_fla",python,selection_mouse
+154,581870,"jasmine/utils/nn.py",9971,14,"self.use_flash",python,selection_mouse
+155,581886,"jasmine/utils/nn.py",9971,15,"self.use_flash_",python,selection_mouse
+156,581904,"jasmine/utils/nn.py",9971,16,"self.use_flash_a",python,selection_mouse
+157,581918,"jasmine/utils/nn.py",9971,17,"self.use_flash_at",python,selection_mouse
+158,581950,"jasmine/utils/nn.py",9971,18,"self.use_flash_att",python,selection_mouse
+159,581984,"jasmine/utils/nn.py",9971,19,"self.use_flash_atte",python,selection_mouse
+160,582035,"jasmine/utils/nn.py",9971,20,"self.use_flash_atten",python,selection_mouse
+161,582173,"jasmine/utils/nn.py",9971,21,"self.use_flash_attent",python,selection_mouse
+162,582305,"jasmine/utils/nn.py",9971,22,"self.use_flash_attenti",python,selection_mouse
+163,582370,"jasmine/utils/nn.py",9971,23,"self.use_flash_attentio",python,selection_mouse
+164,582454,"jasmine/utils/nn.py",9971,24,"self.use_flash_attention",python,selection_mouse
+165,583104,"jasmine/utils/nn.py",9971,24,"",python,content
+166,584569,"jasmine/utils/nn.py",9971,0,"use_flash_attention=False",python,content
+167,597731,"jasmine/utils/nn.py",3266,0,"",python,selection_mouse
+168,598237,"jasmine/utils/nn.py",3266,5,"self.",python,selection_mouse
+169,598238,"jasmine/utils/nn.py",3266,19,"self.use_flash_atte",python,selection_mouse
+170,598238,"jasmine/utils/nn.py",3266,20,"self.use_flash_atten",python,selection_mouse
+171,598256,"jasmine/utils/nn.py",3266,21,"self.use_flash_attent",python,selection_mouse
+172,598333,"jasmine/utils/nn.py",3266,22,"self.use_flash_attenti",python,selection_mouse
+173,598536,"jasmine/utils/nn.py",3266,23,"self.use_flash_attentio",python,selection_mouse
+174,598567,"jasmine/utils/nn.py",3266,24,"self.use_flash_attention",python,selection_mouse
+175,600167,"jasmine/utils/nn.py",3290,0,"=False",python,content
+176,600167,"jasmine/utils/nn.py",3266,5,"",python,content
+177,602827,"jasmine/utils/nn.py",3419,0,"",python,selection_mouse
+178,605557,"jasmine/utils/nn.py",0,0,"",python,tab
+179,608088,"jasmine/utils/nn.py",3273,0,"",python,selection_mouse
+180,608876,"jasmine/utils/nn.py",3290,0,"",python,selection_mouse
+181,609034,"jasmine/utils/nn.py",3286,5,"False",python,selection_mouse
+182,609431,"jasmine/utils/nn.py",3281,0,"",python,selection_mouse
+183,610011,"jasmine/utils/nn.py",3237,0,"",python,selection_mouse
+184,610125,"jasmine/utils/nn.py",17441,0,"",python,selection_command
+185,615407,"TERMINAL",0,0,"git diff",,terminal_command
+186,615440,"TERMINAL",0,0,"]633;C[?1h=\rdiff --git a/jasmine/utils/nn.py b/jasmine/utils/nn.py\r\nindex 473e6be..dce4fb5 100644\r\n--- a/jasmine/utils/nn.py\r\n+++ b/jasmine/utils/nn.py\r\n@@ -101,7 +101,7 @@ class STBlock(nnx.Module):\r\n             param_dtype=self.param_dtype,\r\n             dtype=self.dtype,\r\n             attention_fn=_create_flash_attention_fn(\r\n-                self.use_flash_attention, is_causal=True\r\n+                use_flash_attention=False, is_causal=True\r\n             ),\r\n             rngs=rngs,\r\n             decode=False,\r\n@@ -312,7 +312,7 @@ class TransformerBlock(nnx.Module):\r\n             param_dtype=self.param_dtype,\r\n             dtype=self.dtype,\r\n             attention_fn=_create_flash_attention_fn(\r\n-                self.use_flash_attention, is_causal=True\r\n+                use_flash_attention=False, is_causal=True\r\n             ),\r\n             rngs=rngs,\r\n             decode=self.decode,\r\n\r[?1l>]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output
+187,627512,"TERMINAL",0,0,"idling",,terminal_command
+188,627617,"TERMINAL",0,0,"]633;C[?1049h(B[?7hEvery 1.0s: sinfo_t_idlehkn1990.localdomain: Sun Oct 19 12:07:56 2025Partition dev_cpuonly:     11 nodes idle\rPartition cpuonly:    113 nodes idle\rPartition dev_accelerated:\t 3 nodes idle\rPartition accelerated:     65 nodes idle\rPartition dev_accelerated-h100    :\t 0 nodes idle\rPartition accelerated-h100:\t 0 nodes idle\rPartition large:\t 7 nodes idle\rPartition accelerated-h200:\t 5 nodes idle",,terminal_output
+189,628641,"TERMINAL",0,0,"7",,terminal_output
+190,629681,"TERMINAL",0,0,"8",,terminal_output
+191,630717,"TERMINAL",0,0,"94",,terminal_output
+192,631752,"TERMINAL",0,0,"8:00",,terminal_output
+193,632793,"TERMINAL",0,0,"1",,terminal_output
+194,633839,"TERMINAL",0,0,"2",,terminal_output
+195,634949,"TERMINAL",0,0,"3",,terminal_output
+196,635922,"TERMINAL",0,0,"4",,terminal_output
+197,636964,"TERMINAL",0,0,"5",,terminal_output
+198,638002,"TERMINAL",0,0,"7",,terminal_output
+199,639038,"TERMINAL",0,0,"8",,terminal_output
+200,640076,"TERMINAL",0,0,"9",,terminal_output
+201,641119,"TERMINAL",0,0,"10",,terminal_output
+202,641187,"TERMINAL",0,0,"[?1049l\r[?1l>]0;tum_cte0515@hkn1990:~/Projects/jasmine",,terminal_output