Instructions to use abacaj/starcoderbase-1b-sft with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use abacaj/starcoderbase-1b-sft with Transformers:

# Use a pipeline as a high-level helper
from transformers import pipeline

pipe = pipeline("text-generation", model="abacaj/starcoderbase-1b-sft")

# Load model directly
from transformers import AutoTokenizer, AutoModelForCausalLM

tokenizer = AutoTokenizer.from_pretrained("abacaj/starcoderbase-1b-sft")
model = AutoModelForCausalLM.from_pretrained("abacaj/starcoderbase-1b-sft")

Notebooks
Google Colab
Kaggle
Local Apps

vLLM

How to use abacaj/starcoderbase-1b-sft with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "abacaj/starcoderbase-1b-sft"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "abacaj/starcoderbase-1b-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker

docker model run hf.co/abacaj/starcoderbase-1b-sft

SGLang

How to use abacaj/starcoderbase-1b-sft with SGLang:

Install from pip and serve model

# Install SGLang from pip:
pip install sglang
# Start the SGLang server:
python3 -m sglang.launch_server \
    --model-path "abacaj/starcoderbase-1b-sft" \
    --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": "abacaj/starcoderbase-1b-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Use Docker images

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 "abacaj/starcoderbase-1b-sft" \
        --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": "abacaj/starcoderbase-1b-sft",
		"prompt": "Once upon a time,",
		"max_tokens": 512,
		"temperature": 0.5
	}'

Docker Model Runner
How to use abacaj/starcoderbase-1b-sft with Docker Model Runner:
```
docker model run hf.co/abacaj/starcoderbase-1b-sft
```

starcoderbase-1b-sft / README.md

abacaj

Update README.md

e7f74c2 over 2 years ago

preview code

raw

history blame contribute delete

2.22 kB

metadata

datasets:
  - theblackcat102/evol-codealpaca-v1
model-index:
  - name: abacaj/starcoderbase-1b-sft
    results:
      - task:
          type: text-generation
        dataset:
          type: openai_humaneval
          name: HumanEval
        metrics:
          - name: pass@1
            type: pass@1
            value: 39
            verified: false
      - task:
          type: text-generation
        dataset:
          type: mbpp
          name: MBPP
        metrics:
          - name: pass@1
            type: pass@1
            value: 31.74
            verified: false
language:
  - en

Dataset credits go to: theblackcat102

How to run inference:

import transformers
import torch


def fmt_prompt(prompt: str) -> str:
    return f"""[Instructions]:\n{prompt}\n\n[Response]:"""


if __name__ == "__main__":
    model_name = "abacaj/starcoderbase-1b-sft"
    tokenizer = transformers.AutoTokenizer.from_pretrained(model_name)

    model = (
        transformers.AutoModelForCausalLM.from_pretrained(
            model_name,
        )
        .to("cuda:0")
        .eval()
    )

    prompt = "Write a python function to sort the following array in ascending order, don't use any built in sorting methods: [9,2,8,1,5]"
    prompt_input = fmt_prompt(prompt)
    inputs = tokenizer(prompt_input, return_tensors="pt").to(model.device)
    input_ids_cutoff = inputs.input_ids.size(dim=1)

    with torch.no_grad():
        generated_ids = model.generate(
            **inputs,
            use_cache=True,
            max_new_tokens=512,
            temperature=0.2,
            top_p=0.95,
            do_sample=True,
            eos_token_id=tokenizer.eos_token_id,
            pad_token_id=tokenizer.pad_token_id,
        )

    completion = tokenizer.decode(
        generated_ids[0][input_ids_cutoff:],
        skip_special_tokens=True,
    )

    print(completion)

Evals:

Training charts:

Link to charts: https://api.wandb.ai/links/abacaj1/c4nkcs9r

Code to train model: https://github.com/abacaj/train-with-fsdp