微调sensevoice模型Update best acc: 0.0000

[xiulianzw]

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）

epoch: 2/100, data_slice: 0/1, step_in_slice: 879/4575, step_in_epoch: 879, total step: 10030, (loss_avg_rank: -0.188), (loss_avg_slice: 15.580), (ppl_avg_slice: 5.839e+06), (acc_avg_slice: 0.000), (lr: 1.605e-04), [('loss_ctc', -0.192), ('loss_rich', 0.004), ('loss', -0.188), ('acc_rich', 1.0)], {'data_load': '0.001', 'forward_time': '0.165', 'backward_time': '0.215', 'optim_time': '0.207', 'total_time': '0.599'}, GPU, memory: usage: 3.706 GB, peak: 11.380 GB, cache: 11.947 GB, cache_peak: 12.174 GB

❓ Questions and Help

1. 使用脚本微调iic/SenseVoiceSmall模型的时候，Update best acc: 0.0000
2. 训练的时候为啥loss还会出现负数？acc_rich是训练的准确率吗？

Before asking:

1. search the issues.
2. search the docs.

What is your question?

Code

What have you tried?

What's your environment?

• OS (e.g., Linux): CentOS
• FunASR Version (e.g., 1.0.0): 1.1.5
• ModelScope Version (e.g., 1.11.0): 1.17.1
• PyTorch Version (e.g., 2.0.0): 2.1.0
• How you installed funasr (pip, source): pip
• Python version: 3.8.19
• GPU (e.g., V100M32)：3090RTX
• CUDA/cuDNN version (e.g., cuda11.7): 11.8
• Any other relevant information: