This is a fine-tuned version of Qwen3 4B using one reasoning and one non-reasoning dataset from closed-source LLMs (made available by reedmayhew, thanks!).
The total size of this training dataset is around 300 rows. This model was fine-tuned for 3000 steps.
