diff --git a/tests/entrypoints/multimodal/openai/chat_completion/test_chat_completion_with_mixed_audio_embeds.py b/tests/entrypoints/multimodal/openai/chat_completion/test_chat_completion_with_mixed_audio_embeds.py index fd66d2d57f..a417f860f6 100644 --- a/tests/entrypoints/multimodal/openai/chat_completion/test_chat_completion_with_mixed_audio_embeds.py +++ b/tests/entrypoints/multimodal/openai/chat_completion/test_chat_completion_with_mixed_audio_embeds.py @@ -17,6 +17,7 @@ from transformers import AutoConfig, AutoTokenizer from tests.utils import RemoteOpenAIServer from vllm.utils.serial_utils import tensor2base64 +from vllm.utils.torch_utils import is_torch_equal_or_newer QWEN2AUDIO_MODEL = "Qwen/Qwen2-Audio-7B-Instruct" @@ -148,6 +149,7 @@ def qwen2audio_aligned_content_and_embeds_b64() -> tuple[str, str]: False, id="text-then-audio_embeds", marks=pytest.mark.xfail( + condition=is_torch_equal_or_newer("2.12.0"), reason="torch 2.12 regression: prompt_embeds output diverges " "from raw-text when text precedes audio; " "https://github.com/pytorch/pytorch/issues/184431",