[ROCm][CI] Fix XPASS(strict) on mixed audio embeds test (#49733)
Signed-off-by: Djordje Ramic <djoramic@amd.com> Signed-off-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: Andreas Karatzas <akaratza@amd.com> Co-authored-by: OpenAI Codex <codex@openai.com>
djramic committed
Jul 24, 2026 at 23:09 UTC
7513d071bd749c938181fac9e67ac103d0c52dc1
1 file changed
+2
tests/entrypoints/multimodal/openai/chat_completion/test_chat_completion_with_mixed_audio_embeds.py
+2
@@ -17,6 +17,7 @@ from transformers import AutoConfig, AutoTokenizer
17
18
from tests.utils import RemoteOpenAIServer
19
from vllm.utils.serial_utils import tensor2base64
20
+from vllm.utils.torch_utils import is_torch_equal_or_newer
21
22
QWEN2AUDIO_MODEL = "Qwen/Qwen2-Audio-7B-Instruct"
23
@@ -148,6 +149,7 @@ def qwen2audio_aligned_content_and_embeds_b64() -> tuple[str, str]:
149
False,
150
id="text-then-audio_embeds",
151
marks=pytest.mark.xfail(
152
+ condition=is_torch_equal_or_newer("2.12.0"),
153
reason="torch 2.12 regression: prompt_embeds output diverges "
154
"from raw-text when text precedes audio; "
155
"https://github.com/pytorch/pytorch/issues/184431",