104  Downloads Updated  1 month ago

Fine-tuned Qwen3.5-9B with distilled reasoning from research-backed datasets. Trained via LoRA SFT with an additive data strategy that preserves base model capabilities while improving instruction following and reasoning.