transformers icon indicating copy to clipboard operation
transformers copied to clipboard

align xpu's autocast behavior w/ cuda by using device agnostic torch APIs

Open yao-matrix opened this issue 6 months ago • 6 comments

@ArthurZucker, pls help review, thx very much.

yao-matrix avatar May 22 '25 07:05 yao-matrix

cc @IlyasMoutawwakil

Rocketknight1 avatar May 22 '25 16:05 Rocketknight1

ci failure seems not brought by my PR

yao-matrix avatar May 23 '25 01:05 yao-matrix

@ArthurZucker @IlyasMoutawwakil could you help review and comment? Thx very much

yao-matrix avatar May 26 '25 22:05 yao-matrix

ci failure maybe because of the instable ci env

yao-matrix avatar May 29 '25 06:05 yao-matrix

CI seems clear now! cc @IlyasMoutawwakil

Rocketknight1 avatar Jun 04 '25 12:06 Rocketknight1

@IlyasMoutawwakil , could you help review? Thx very much.

yao-matrix avatar Jun 09 '25 01:06 yao-matrix

@Rocketknight1 , do you know who else need review this PR after Ilyas approved? Thx.

yao-matrix avatar Jun 18 '25 23:06 yao-matrix

run-slow: qwen2_5_omni, gemma, phimoe, qwen2_moe, gpt2, distilbert

ydshieh avatar Jun 19 '25 11:06 ydshieh

This comment contains run-slow, running the specified jobs:

models: ['models/distilbert', 'models/gemma', 'models/gpt2', 'models/phimoe', 'models/qwen2_5_omni', 'models/qwen2_moe'] quantizations: [] ...

github-actions[bot] avatar Jun 19 '25 11:06 github-actions[bot]

The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update.