FHL1998

TypeError: getattr(): attribute name must be string

1

Traceback (most recent call last): File "qlora.py", line 791, in train() File "qlora.py", line 636, in train tokenizer = AutoTokenizer.from_pretrained( File "/opt/conda/envs/env/lib/python3.8/site-packages/transformers/models/auto/tokenization_auto.py", line 635, in from_pretrained tokenizer_class = tokenizer_class_from_name(tokenizer_class_name) File...

Evaluation loss issue (strange trend)

3

A weird thing happens when I inspect the val_loss, it decreases and then increases in a strange way. Does anyone have an idea of solving it? 

[Need Help] Setting train_on_source=True achieved less train_loss

1

I find a strange thing that when I set `train_on_source=Ture`, the train loss is smaller. When I set `train_on_source=False`, the evaluation will output something from the source, it seems IGNORE_INDEX...