1. Add example for other model (DeepScaleR-1.5B-Preview) and dataset in docs/guides/eval.md.
   (Default eval config is test AIME on Qwen2.5-Math-1.5B-Instruct)
2. Test and add eval link in the training guides: docs/guides/dpo|grpo|sft.md.
   I think there is no need to add convert guide in the training guides since it is already in eval guide.
3. Support Pass@1 accuracy averaged over n samples.

Link docs/guides/eval.md on the front page readme after all these done.