starcoder2-15b-instruct-v0.1/README.md

pipeline_tag, base_model, datasets, license, library_name, tags, model-index

pipeline_tag	base_model	datasets	license	library_name	tags	model-index
text-generation	bigcode/starcoder2-15b	bigcode/self-oss-instruct-sc2-exec-filter-50k	bigcode-openrail-m	transformers	code	name results starcoder2-15b-instruct-v0.1 task dataset metrics type text-generation name type LiveCodeBench (code generation) livecodebench-codegeneration type value pass@1 20.4 task dataset metrics type text-generation name type LiveCodeBench (self repair) livecodebench-selfrepair type value pass@1 20.9 task dataset metrics type text-generation name type LiveCodeBench (test output prediction) livecodebench-testoutputprediction type value pass@1 29.8 task dataset metrics type text-generation name type LiveCodeBench (code execution) livecodebench-codeexecution type value pass@1 28.1 task dataset metrics type text-generation name type HumanEval humaneval type value pass@1 72.6 task dataset metrics type text-generation name type HumanEval+ humanevalplus type value pass@1 63.4 task dataset metrics type text-generation name type MBPP mbpp type value pass@1 75.2 task dataset metrics type text-generation name type MBPP+ mbppplus type value pass@1 61.2 task dataset metrics type text-generation name type DS-1000 ds-1000 type value pass@1 40.6

StarCoder2-Instruct: Fully Transparent and Permissive Self-Alignment for Code Generation

Model Summary

We introduce StarCoder2-15B-Instruct-v0.1, the very first entirely self-aligned code Large Language Model (LLM) trained with a fully permissive and transparent pipeline. Our open-source pipeline uses StarCoder2-15B to generate thousands of instruction-response pairs, which are then used to fine-tune StarCoder-15B itself without any human annotations or distilled data from huge and proprietary LLMs.

• Model: bigcode/starcoder2-15b-instruct-v0.1
• Code: bigcode-project/starcoder2-self-align
• Dataset: bigcode/self-oss-instruct-sc2-exec-filter-50k
• Authors: Yuxiang Wei, Federico Cassano, Jiawei Liu, Yifeng Ding, Naman Jain, Harm de Vries, Leandro von Werra, Arjun Guha, Lingming Zhang.

Citation

@article{wei2024selfcodealign,
  title={SelfCodeAlign: Self-Alignment for Code Generation}, 
  author={Yuxiang Wei and Federico Cassano and Jiawei Liu and Yifeng Ding and Naman Jain and Zachary Mueller and Harm de Vries and Leandro von Werra and Arjun Guha and Lingming Zhang},
  year={2024},
  journal={arXiv preprint arXiv:2410.24198}
}

Use

Intended use

The model is designed to respond to coding-related instructions in a single turn. Instructions in other styles may result in less accurate responses.

Here is an example to get started with the model using the transformers library:

import transformers
import torch

pipeline = transformers.pipeline(
    model="bigcode/starcoder2-15b-instruct-v0.1",
    task="text-generation",
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

def respond(instruction: str, response_prefix: str) -> str:
    messages = [{"role": "user", "content": instruction}]
    prompt = pipeline.tokenizer.apply_chat_template(messages, tokenize=False)
    prompt += response_prefix

    teminators = [
        pipeline.tokenizer.eos_token_id,
        pipeline.tokenizer.convert_tokens_to_ids("###"),
    ]

    result = pipeline(
        prompt,
        max_length=256,
        num_return_sequences=1,
        do_sample=False,
        eos_token_id=teminators,
        pad_token_id=pipeline.tokenizer.eos_token_id,
        truncation=True,
    )
    response = response_prefix + result[0]["generated_text"][len(prompt) :].split("###")[0].rstrip()
    return response


instruction = "Write a quicksort function in Python with type hints and a 'less_than' parameter for custom sorting criteria."
response_prefix = ""

print(respond(instruction, response_prefix))

Here is the expected output:

Here's how you can implement a quicksort function in Python with type hints and a 'less_than' parameter for custom sorting criteria:

```python
from typing import TypeVar, Callable

T = TypeVar('T')

def quicksort(items: list[T], less_than: Callable[[T, T], bool] = lambda x, y: x < y) -> list[T]:
    if len(items) <= 1:
        return items

    pivot = items[0]
    less = [x for x in items[1:] if less_than(x, pivot)]
    greater = [x for x in items[1:] if not less_than(x, pivot)]
    return quicksort(less, less_than) + [pivot] + quicksort(greater, less_than)
```

Bias, Risks, and Limitations

StarCoder2-15B-Instruct-v0.1 is primarily finetuned for Python code generation tasks that can be verified through execution, which may lead to certain biases and limitations. For example, the model might not adhere strictly to instructions that dictate the output format. In these situations, it's beneficial to provide a response prefix or a one-shot example to steer the model’s output. Additionally, the model may have limitations with other programming languages and out-of-domain coding tasks.

The model also inherits the bias, risks, and limitations from its base StarCoder2-15B model. For more information, please refer to the StarCoder2-15B model card.

Evaluation on EvalPlus, LiveCodeBench, and DS-1000

Training Details

Hyperparameters

• Optimizer: Adafactor
• Learning rate: 1e-5
• Epoch: 4
• Batch size: 64
• Warmup ratio: 0.05
• Scheduler: Linear
• Sequence length: 1280
• Dropout: Not applied

Hardware

1 x NVIDIA A100 80GB

Resources

• Model: bigcode/starCoder2-15b-instruct-v0.1
• Code: bigcode-project/starcoder2-self-align
• Dataset: bigcode/self-oss-instruct-sc2-exec-filter-50k

Full Data Pipeline

Our dataset generation pipeline has several steps. We provide intermediate datasets for every step of the pipeline:

1. Original seed dataset filtered from The Stack v1: https://huggingface.co/datasets/bigcode/python-stack-v1-functions-filtered
2. Seed dataset filtered using StarCoder2-15B as a judge for removing items with bad docstrings: https://huggingface.co/datasets/bigcode/python-stack-v1-functions-filtered-sc2
3. seed -> concepts: https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-concepts
4. concepts -> instructions: https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-instructions
5. instructions -> response: https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-responses-unfiltered
6. Responses filtered by executing them: https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-500k-raw
7. Executed responses filtered by deduplicating them (final dataset): https://huggingface.co/datasets/bigcode/self-oss-instruct-sc2-exec-filter-50k