archit11/track_b_sft

Track B SFT — Python Coding Instructions: RL dataset on Hugging Face. A synthetic instruction-tuning dataset for Python coding tasks, generated from the verl open-source RL framework codebase. Used to fine-tune Qwen/Qwen2.5-Coder-1.5B with LoRA, achieving +24 point absolute improvement in pass@1…