ColossalAI Update python

fix trace error when batchsize eq 1 and the model is clip model

Oct 27 '22 12:10 tangbogreat

test_modeling_clip.py has been added in the second commit. You can review and run it to reproduce !

Oct 29 '22 08:10 tangbogreat

thx，I will change this code.

发自我的iPhone

------------------ Original ------------------ From: Super Daniel @.> Date: Mon,Oct 31,2022 2:41 PM To: hpcaitech/ColossalAI @.> Cc: Bo tang @.>, Author @.> Subject: Re: [hpcaitech/ColossalAI] Update python_ops.py (PR #1770)

@super-dainiu commented on this pull request.

In tests/test_auto_parallel/test_tensor_shard/test_modeling_clip.py: > +model = CLIPModel.from_pretrained("openai/clip-vit-base-patch32") +model = model.to('meta') +processor = CLIPProcessor.from_pretrained("openai/clip-vit-base-patch32")
don't load pretrained weights, cuz for the CI testing it will spend a lot of time. Can refer to these files. https://github.com/hpcaitech/ColossalAI/tree/main/tests/test_fx/test_tracer/test_hf_model

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

Oct 31 '22 13:10 tangbogreat

Does this PR still valid? Or the developer has gavin up?

Jan 07 '23 15:01 feifeibear

Closed due to inactivity.

Jan 29 '23 05:01 binmakeswell