robin-linzmayer/Qwen3B-s1k-gemini-sft-exp0

Create training_config_messy

Upload s1k-3b-loss-curves.png

Add tokenizer

Trained on gemini_thinking_trajectory from s1k-compressed dataset

initial commit