Ronghang Hu comments

Results 46 comments of


Ronghang Hu

Train UniT with dataset other than 8 datasets specified in tutorial

Hi, since Hateful Memes is not used for detection tasks, it should not be added to `model_config.unit.base_args.num_queries.detection` in "B: Config file". Instead, it should be added only under `model_config.unit.base_args.num_queries.vl`, as...

Pretrained models cannot replicate the experiment results in paper under run_type Val

Hi @xzChiang, this seems weird. I just tried running this evaluation locally on a two-GPU machine, and got 0.4065 accuracy ``` mmf_run config=projects/m4c/configs/textvqa/joint_with_stvqa.yaml \ datasets=textvqa \ model=m4c \ run_type=val \...

Pretrained models cannot replicate the experiment results in paper under run_type Val

> I get the same problem. How to solve this ? Is there any configures or settings should be specially noticed or modified ? @akira-l Did you download the TextVQA...

Pretrained models cannot replicate the experiment results in paper under run_type Val

@xzChiang @akira-l The warnings above should be fine and should not affect training or evaluation. (They are raised because the M4C model checkpoint is saved from an earlier Pythia/MMF version,...

Pretrained models cannot replicate the experiment results in paper under run_type Val

@akira-l thanks for the update. Unfortunately I still could not reproduce this error on my end. Could you share your PyTorch environment (the output of `python -m torch.utils.collect_env`)? Since your...

Autograd discrepancy in `nn.Linear` (`torch.nn.functional.linear`) between native PyTorch and PyTorch/XLA

> Hi @ronghanghu, I wanted to confirm whether the autograd discrepancy in `nn.Linear()` is specific to FSDP or applies to all instances of `nn.Linear()` when using PyTorch/XLA. cc @JackCaoG Hi...