---
language:
- en
- zh
license: other
tags:
- gensyn
- testnet
- rl-swarm
- code-generation
- mbpp
- code-contests
pipeline_tag: text-generation
base_model: Qwen/Qwen2.5-Coder-0.5B-Instruct
datasets:
- deepmind/code_contests
- google-research-datasets/mbpp
library_name: transformers
---

# Qwen2.5-Coder-0.5B-Instruct — Gensyn Swarm

此仓库用于记录在 Gensyn Testnet 的 RL Swarm 代码生成任务中的参与与权重版本。基础模型为 `Qwen/Qwen2.5-Coder-0.5B-Instruct`。

## Overview
- Base model: Qwen2.5-Coder-0.5B-Instruct
- Task: Code generation (mbpp, code_contests)
- Hardware: Mac mini (M4, 16GB), Apple MPS
- Participation: Gensyn Testnet / CodeZero

## Training Data
- deepmind/code_contests
- google-research-datasets/mbpp

## Metrics
- pass@1, pass@k, exact-match（随版本更新补充）

## Usage
```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "aiyun123/Qwen2.5-Coder-0.5B-Instruct"
tok = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto"  # 自动选择设备；在 Mac 可走 MPS
)

prompt = "Write a Python function to check if a number is prime."
inputs = tok(prompt, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=256)
print(tok.decode(outputs[0], skip_special_tokens=True))
```

### Inference Notes
- macOS: 推荐 `PYTORCH_ENABLE_MPS_FALLBACK=1`，必要时设置 `PYTORCH_MPS_HIGH_WATERMARK_RATIO=0.0`
- dtype: 若遇内存限制，可尝试 `torch.float16` 或启用 `device_map="auto"`

## Limitations
- 小模型在复杂算法/长代码生成上的能力有限；需结合评测任务客观比较

## Versioning
- `swarm-YYYY-MM-DD`（每日/每轮次版本号；后续推送时更新）

## License
与上游基础模型许可一致；请参见 Qwen2.5 的官方许可说明与链接。