Jason Chou comments

Results 34 comments of


                                            Jason Chou

`hub_utils.py` assumes a different sharding convention

> @EIFY I think this issue might be related to the one #407 you encountered before, where the distributed process groups were not initialized correctly. Have you tried removing `"--distributed-port...

Python version requirement more strict than stated

FYI, [specified branch of Megatron-LM](https://github.com/facebookresearch/metaseq/blob/7828d72815a9a581ab47b95876d38cb262741883/docs/setup.md) ``` git clone --branch fairseq_v2 https://github.com/ngoyal2707/Megatron-LM.git ``` also no longer works with newer versions of PyTorch since the header file it wants to include `#include...

Re-release consolidated OPT / OPT-IML checkpoints

I can work around it, but could the following issue be considered related as well? * [](https://github.com/facebookresearch/metaseq/issues/552) #552

Modifying ALiBi for Encoder-Attention or Cross-Attention

Hi @VarunGumma, it has been a while and I didn't do much work with `transformer_decoder`, but that logic was [already there before my change](https://github.com/EIFY/fairseq/commit/fc1fabc8612cd25cf3e15a5623ebddd59f1219bd#diff-a3296485b7223218deda82ae22f2710ebd711f572088f7538894d50d4580560aL335). I merely moved it out of...