YCWTG/Qwen3-Coder-Next-Open-Code-SFT-ChatML
收藏Hugging Face2026-04-05 更新2026-04-12 收录
下载链接：
https://hf-mirror.com/datasets/YCWTG/Qwen3-Coder-Next-Open-Code-SFT-ChatML
下载链接
链接失效反馈官方服务：
资源简介：
# Qwen3-Coder-Next-Open-Code-SFT-ChatML
Converted from `zake7749/Qwen3-Coder-Next-Open-Code-SFT`.
Files:
- `train.jsonl`
- `validation.jsonl`
Each JSONL row contains exactly one field:
- `messages`
Rules:
- keep only `test_pass_rate == 1.0`
- `problem_id` = `id` without suffix `_sft_<number>`
- if a problem has >= 4 samples: put 1 sample (median by response length) into validation
- otherwise no validation sample for that problem
- train keeps up to 3 samples per problem (shortest / median / longest by response length)
- no `test` split
- thought removes code from `response` and inserts:
```python
Code intentionally omitted from the thought block. The final answer must contain only the completed Python code.
```
提供机构：
YCWTG
