transformers
cb871238 - [SGLang CI] Fix the VLM teardown error on CPU (follow-up of #50913) (#50914)

Commit
2 days ago
[SGLang CI] Fix the VLM teardown error on CPU (follow-up of #50913) (#50914) [SGLang CI] Fix the VLM teardown error on CPU instead of deselecting a test The vision-language error comes from TestQwenVLUnderstandsImage.tearDownClass calling empty_gpu_cache() (torch.accelerator.empty_cache(), which raises without an accelerator), not from test_colliding_low_hash_bits_do_not_reuse_image_kv itself (it passes); pytest reports it on the class's last test. The deselect from #50913 also never matched (pytest's rootdir is sglang/test, so node ids don't start with test/). Revert it and make empty_gpu_cache() a no-op without an accelerator in the conftest the workflow already writes. Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
Author
Parents
Loading