layer-skip-exps
Collection
2 items • Updated
How to use ariG23498/layer-skip-v1 with Transformers:
# Use a pipeline as a high-level helper
from transformers import pipeline
pipe = pipeline("text-generation", model="ariG23498/layer-skip-v1") # Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM
tokenizer = AutoTokenizer.from_pretrained("ariG23498/layer-skip-v1")
model = AutoModelForCausalLM.from_pretrained("ariG23498/layer-skip-v1")How to use ariG23498/layer-skip-v1 with vLLM:
# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "ariG23498/layer-skip-v1"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "ariG23498/layer-skip-v1",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'docker model run hf.co/ariG23498/layer-skip-v1
How to use ariG23498/layer-skip-v1 with SGLang:
# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
--model-path "ariG23498/layer-skip-v1" \
--host 0.0.0.0 \
--port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "ariG23498/layer-skip-v1",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'docker run --gpus all \
--shm-size 32g \
-p 30000:30000 \
-v ~/.cache/huggingface:/root/.cache/huggingface \
--env "HF_TOKEN=<secret>" \
--ipc=host \
lmsysorg/sglang:latest \
python3 -m sglang.launch_server \
--model-path "ariG23498/layer-skip-v1" \
--host 0.0.0.0 \
--port 30000
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:30000/v1/completions" \
-H "Content-Type: application/json" \
--data '{
"model": "ariG23498/layer-skip-v1",
"prompt": "Once upon a time,",
"max_tokens": 512,
"temperature": 0.5
}'How to use ariG23498/layer-skip-v1 with Docker Model Runner:
docker model run hf.co/ariG23498/layer-skip-v1
class LayerSkipSFTTrainer(SFTTrainer):
def __init__(self, *args, **kwargs):
super().__init__(*args, **kwargs)
self.early_exit_layer = 0 # initialize with 0
self.always_last_layer = True
def compute_loss(self, model, inputs, return_outputs=False, num_items_in_batch=None):
self.early_exit_layer = (self.early_exit_layer % (model.config.num_hidden_layers - 1)) + 1 # rotates between [1, num_hidden_layers-1]
labels = inputs.pop("labels")
outputs = model(**inputs, output_hidden_states=True)
hidden_state = outputs["hidden_states"][self.early_exit_layer]
logits = model.lm_head(hidden_state)
loss = model.loss_function(logits=logits, labels=labels, vocab_size=model.vocab_size)
if self.always_last_layer:
loss = loss + model.loss_function(logits=outputs["logits"], labels=labels, vocab_size=model.vocab_size)
return loss
Base model
meta-llama/Llama-3.2-1B