Ya-Qi Yu

Model parameters require gradients during distributed testing

1

**Describe the bug** Transformer-based models can cause CUDA out of memory error when evaluated on multiple GPUs. **Reproduction** Set the max_seq_len to a large number, e.g. 1000, then evaluate a...

test.py is not using CTCLabelConverterForBaiduWarpctc https://github.com/clovaai/deep-text-recognition-benchmark/blob/68a80fe97943a111ff1efaf52a63ad8f0f1c0e5d/test.py#L192-L195 which conflicts with https://github.com/clovaai/deep-text-recognition-benchmark/blob/68a80fe97943a111ff1efaf52a63ad8f0f1c0e5d/test.py#L113-L115 using CTCLabelConverterForBaiduWarpctc can fix it https://github.com/clovaai/deep-text-recognition-benchmark/blob/68a80fe97943a111ff1efaf52a63ad8f0f1c0e5d/train.py#L47-L53

About the tokenizer

2

Why the tokenizer class specified in tokenizer config is BloomTokenizer? There is only BloomTokenizerFast in transformers implementation.

RuntimeError: expected scalar type Half but found Float

26

I met the following error when I tried to train bloom-7b1-mt with peft LoRA in 8bit+fp16 (torch amp) mode: ```sh Traceback (most recent call last): File "finetune.py", line 141, in...

Canonicalization of column header

3

Hi, thanks for releasing the PubTables1M dataset. It took me a lot of time to clean the PubTabNet dataset, and the oversegmentation problem is probably the most tricky part. The...