README.md

---
license: other
language:
- en
tags:
- causal-lm
- code
metrics:
- code_eval
library_name: transformers
model-index:
- name: dgtalbug/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
  - task:
      type: text-generation
    dataset:
      type: nuprl/MultiPL-E
      name: MultiPL-HumanEval (C++)
    metrics:
    - name: pass@1
      type: pass@1
      value: 30.9
  - task:
      type: text-generation
    dataset:
      type: nuprl/MultiPL-E
      name: MultiPL-HumanEval (Java)
    metrics:
    - name: pass@1
      type: pass@1
      value: 32.1
  - task:
      type: text-generation
    dataset:
      type: nuprl/MultiPL-E
      name: MultiPL-HumanEval (JavaScript)
    metrics:
    - name: pass@1
      type: pass@1
      value: 32.1
  - task:
      type: text-generation
    dataset:
      type: nuprl/MultiPL-E
      name: MultiPL-HumanEval (PHP)
    metrics:
    - name: pass@1
      type: pass@1
      value: 24.2
  - task:
      type: text-generation
    dataset:
      type: nuprl/MultiPL-E
      name: MultiPL-HumanEval (Rust)
    metrics:
    - name: pass@1
      type: pass@1
      value: 23.0
---

# **Stable Code Instruct 3B — Base Model**

> This repository stores an **unchanged** copy of `stabilityai/stable-code-instruct-3b`  
> for use as a **base model** in future fine‑tuning projects (including Stephen).

---

## 📌 About the Model

`stable-code-instruct-3b` is a **2.7B parameter decoder-only transformer** from Stability AI, tuned for multi‑language code generation and conversational coding assistance.  
It is suitable as a **starting point** for specialized code assistants,  
including fine‑tuned variants with domain‑specific datasets.

**Key Features:**
- General purpose code generation across multiple programming languages.
- Instruction‑tuned for better conversational performance.
- Strong performance on [MultiPL-E](https://github.com/nuprl/MultiPL-E) benchmarks.

---

## 📊 Performance (MultiPL-E Benchmark)

| Language     | pass@1 |
|--------------|--------|
| Python       | 32.4%  |
| C++          | 30.9%  |
| Java         | 32.1%  |
| JavaScript   | 32.1%  |
| PHP          | 24.2%  |
| Rust         | 23.0%  |

---

## 🚀 Usage

```python
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer

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

messages = [
    {"role": "system", "content": "You are a helpful coding assistant."},
    {"role": "user", "content": "Write a Python function to reverse a string."}
]

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=200,
    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=True)[0]
print(output)
```

---

## 📜 License

This model follows the **[Stability AI Community License](https://huggingface.co/stabilityai/stable-code-instruct-3b/blob/main/LICENSE.md)**.  
For commercial use, refer to [Stability AI licensing terms](https://stability.ai/license).

---

## 📌 Note for Fine‑Tuning

This repository is **not modified** — it is kept as a **clean base model** for derivative works.  
Fine‑tuned versions (e.g., Stephen) will be released in **separate repositories**.