schedule-free adamw: Analyze why jax performed better than pytorch on finewebedu workload

schedule free adamw:
<img width="1800" height="1050" alt="Image" src="https://github.com/user-attachments/assets/322ff62e-07e2-436c-9e92-47d85a218791" />

adamw:
<img width="1800" height="1050" alt="Image" src="https://github.com/user-attachments/assets/c7df872f-312a-481c-9162-9d79dab88a86" />