{'loss': 4.3273, 'learning_rate': 0.0, 'epoch': 0.13}

[chenzk1993]

With the default parameters, the learning rate becomes 0?

[victorzhz111]

I encountered the same problem. Try adjust the micro batch size with larger number.

[LiuPearl1]

@chenzk1993 the learning rate is not 0, it's a very small value. One decimal place is reserved and the specific value cannot be seen.

[lyccyl1]

This may be due to hardware reasons. On some hardware, the quantization model is not compatible with fp16. You can try set fp16=False.