YCWTG/Qwen3-Coder-Next-Open-Code-SFT-ChatML

# 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. ```