Orbax to huggingface conversion does not support half precision #3489
Description
Bug report
The old conversion script llama_mistral_mixtral_orbax_to_hf.py saved checkpoints in half precision.
The new checkpoint conversion script to_huggingface.py saves chackpoints in full precision which takes more disk storage space.
Logs/Output
No response
Environment Information
No response
Additional Context
No response