metadata

license: other
language:
  - en
tags:
  - causal-lm
  - code
metrics:
  - code_eval
library_name: transformers
model-index:
  - name: stabilityai/stable-code-instruct-3b
    results:
      - task:
          type: text-generation
        dataset:
          type: nuprl/MultiPL-E
          name: MultiPL-HumanEval (Python)
        metrics:
          - name: pass@1
            type: pass@1
            value: 32.4
            verified: false
      - task:
          type: text-generation
        dataset:
          type: nuprl/MultiPL-E
          name: MultiPL-HumanEval (C++)
        metrics:
          - name: pass@1
            type: pass@1
            value: 30.9
            verified: false
      - task:
          type: text-generation
        dataset:
          type: nuprl/MultiPL-E
          name: MultiPL-HumanEval (Java)
        metrics:
          - name: pass@1
            type: pass@1
            value: 32.1
            verified: false
      - task:
          type: text-generation
        dataset:
          type: nuprl/MultiPL-E
          name: MultiPL-HumanEval (JavaScript)
        metrics:
          - name: pass@1
            type: pass@1
            value: 32.1
            verified: false
      - task:
          type: text-generation
        dataset:
          type: nuprl/MultiPL-E
          name: MultiPL-HumanEval (PHP)
        metrics:
          - name: pass@1
            type: pass@1
            value: 24.2
            verified: false
      - task:
          type: text-generation
        dataset:
          type: nuprl/MultiPL-E
          name: MultiPL-HumanEval (Rust)
        metrics:
          - name: pass@1
            type: pass@1
            value: 23
            verified: false

GPTQ quantized version of stable-code-instruct-3b model.

Stable Code Instruct 3B

Try it out here: https://huggingface.co/spaces/stabilityai/stable-code-instruct-3b

Model Description

stable-code-instruct-3b is a 2.7B billion parameter decoder-only language model tuned from stable-code-3b. This model was trained on a mix of publicly available datasets, synthetic datasets using Direct Preference Optimization (DPO).

This instruct tune demonstrates state-of-the-art performance (compared to models of similar size) on the MultiPL-E metrics across multiple programming languages tested using BigCode's Evaluation Harness, and on the code portions of MT Bench. The model is finetuned to make it useable in tasks like,

General purpose Code/Software Engineering like conversations.
SQL related generation and conversation.

Usage

Here's how you can run the model use the model:


import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
tokenizer = AutoTokenizer.from_pretrained("stabilityai/stable-code-instruct-3b", trust_remote_code=True)
model = AutoModelForCausalLM.from_pretrained("stabilityai/stable-code-instruct-3b", torch_dtype=torch.bfloat16, trust_remote_code=True)
model.eval()
model = model.cuda()

messages = [
    {
        "role": "system",
        "content": "You are a helpful and polite assistant",
    },
    {
        "role": "user",
        "content": "Write a simple website in HTML. When a user clicks the button, it shows a random joke from a list of 4 jokes."
    },
]

prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True, tokenize=False)

inputs = tokenizer([prompt], return_tensors="pt").to(model.device)

tokens = model.generate(
    **inputs,
    max_new_tokens=1024,
    temperature=0.5,
    top_p=0.95,
    top_k=100,
    do_sample=True,
    use_cache=True
)

output = tokenizer.batch_decode(tokens[:, inputs.input_ids.shape[-1]:], skip_special_tokens=False)[0]

Model Details

Developed by: Stability AI
Model type: Stable Code Instruct 3B model is an auto-regressive language model based on the transformer decoder architecture.
Language(s): English
Paper: Stable Code Technical Report
Library: Alignment Handbook
Finetuned from model: https://huggingface.co/stabilityai/stable-code-3b
License: StabilityAI Non-Commercial Research Community License. If you want to use this model for your commercial products or purposes, please contact us here to learn more.
Contact: For questions and comments about the model, please email [email protected]

Performance

Multi-PL Benchmark:

Model	Size	Avg	Python	C++	JavaScript	Java	PHP	Rust
Codellama Instruct	7B	0.30	0.33	0.31	0.31	0.29	0.31	0.25
Deepseek Instruct	1.3B	0.44	0.52	0.52	0.41	0.46	0.45	0.28
Stable Code Instruct (SFT)	3B	0.44	0.55	0.45	0.42	0.42	0.44	0.32
Stable Code Instruct (DPO)	3B	0.47	0.59	0.49	0.49	0.44	0.45	0.37

MT-Bench Coding:

Model	Size	Score
DeepSeek Coder	1.3B	4.6
Stable Code Instruct (DPO)	3B	5.8(ours)
Stable Code Instruct (SFT)	3B	5.5
DeepSeek Coder	6.7B	6.9
CodeLlama Instruct	7B	3.55
StarChat2	15B	5.7

SQL Performance

Model	Size	Date	Group By	Order By	Ratio	Join	Where
Stable Code Instruct (DPO)	3B	24.0%	54.2%	68.5%	40.0%	54.2%	42.8%
DeepSeek-Coder Instruct	1.3B	24.0%	37.1%	51.4%	34.3%	45.7%	45.7%
SQLCoder	7B	64.0%	82.9%	74.3%	54.3%	74.3%	74.3%

How to Cite

@misc{stable-code-instruct-3b,
      url={[https://huggingface.co/stabilityai/stable-code-3b](https://huggingface.co/stabilityai/stable-code-instruct-3b)},
      title={Stable Code 3B},
      author={Phung, Duy, and Pinnaparaju, Nikhil and Adithyan, Reshinth and Zhuravinskyi, Maksym and Tow, Jonathan and Cooper, Nathan}
}