Eason Leo comments

Results 6 comments of


                                            Eason Leo

为什么fine-tune过程中loss会忽大忽小呢？

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

[Bug] Training on 13B causes loss to be 0, while 7B works fine

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

fintune 13B model, train_loss always 0.0

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

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

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

词表扩充后具体怎么使用？

U may should resize the embedding layer that the size is right.

TypeError: list indices must be integers or slices, not str

The error traceback: Traceback (most recent call last): File "demo_opt.py", line 470, in eval_results = _evaluate_predictions_on_crowdhuman(gt_path, fpath) File "demo_opt.py", line 448, in _evaluate_predictions_on_crowdhuman database = Database(gt_path, dt_path, target_key, None, mode)...