How is the FP16 model trained?

[icestoneking]

Notice: In order to resolve issues more efficiently, please raise issue following the template.
（注意：为了更加高效率解决您遇到的问题，请按照模板提问，补充细节）

❓How is the FP16 model trained? Can I save the FP16 model as a normal model after training?

Before asking:

What is your question?

1. 用ddp训练:
   ++train_conf.use_fp16=true,最后保存的模型仍为fp32
2. 用deepseed训练:
   FunASR/funasr/models/sanm/attention.py", line 518, in forward
   [rank0]: inputs = inputs * mask
   [rank0]: ~~~~~~~^~~~~~
   [rank0]: RuntimeError: The size of tensor a (0) must match the size of tensor b (74) at non-singleton dimension 1

Code

What have you tried?

What's your environment?

• OS (e.g., Linux):
• FunASR Version (e.g., 1.0.0):
• ModelScope Version (e.g., 1.11.0):
• PyTorch Version (e.g., 2.0.0):
• How you installed funasr (pip, source):
• Python version:
• GPU (e.g., V100M32)
• CUDA/cuDNN version (e.g., cuda12.1):
• Any other relevant information: