Open lcqlalala opened 1 year ago
I also encountered this issue. Have you found any solutions?
Have you tried to add predictions = np.argmax(predictions, axis=-1)
before decoding? Current prediction shape looks like (batch_size, length, vocabulary_size). We should require a shape like (batch_size, length).
Traceback (most recent call last): File "/mnt/ssd/lcq/qlora-main/qlora.py", line 791, in
train()
File "/mnt/ssd/lcq/qlora-main/qlora.py", line 773, in train
predictions = tokenizer.batch_decode(
File "/mnt/ssd/lcq/conda_env/qlora/lib/python3.9/site-packages/transformers/tokenization_utils_base.py", line 3464, in batch_decode
return [
File "/mnt/ssd/lcq/conda_env/qlora/lib/python3.9/site-packages/transformers/tokenization_utils_base.py", line 3465, in
self.decode(
File "/mnt/ssd/lcq/conda_env/qlora/lib/python3.9/site-packages/transformers/tokenization_utils_base.py", line 3504, in decode
return self._decode(
File "/mnt/ssd/lcq/conda_env/qlora/lib/python3.9/site-packages/transformers/tokenization_utils.py", line 931, in _decode
filtered_tokens = self.convert_ids_to_tokens(token_ids, skip_special_tokens=skip_special_tokens)
File "/mnt/ssd/lcq/conda_env/qlora/lib/python3.9/site-packages/transformers/tokenization_utils.py", line 906, in convert_ids_to_tokens
index = int(index)
TypeError: int() argument must be a string, a bytes-like object or a number, not 'list'