--- id: Ggboykxz/t4llm-python-sft license: apache-2.0 pretty_name: t4llm Python SFT (work in progress) tags: - code - python - t4llm --- # t4llm-python-sft — work in progress Crash-only, resumable SFT of `Qwen/Qwen3-1.7B-Base` (Apache-2.0) on Python function-level instruction data (ODC-BY 1.0; see the data repo). Training is done by the [t4llm](https://github.com/Ggboykxz/t4llm) pipeline with crash-safe checkpoints; this repository will hold the merged fp16 export and the f16 GGUF once training completes. No benchmark numbers are published until measured; unmeasured means unmeasured, not zero. **Do not use until training and evaluation are complete.**