iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[AMD][CI] Fix Language Models Test (Extended Generation) failures (#45509)

Browse Source 
Signed-off-by: Oxana Korzh <okorzh@amd.com>
Co-authored-by: Claude <noreply@anthropic.com>
Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
Oxana Korzh
2026-06-19 12:06:56 +08:00
committed by GitHub
co-authored by Claude Cursor
parent c9135db27c
commit ecf9d83520
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/models/language/generation/test_common.py
+5 -1
View File
@@ -130,8 +130,12 @@ def test_models(
monkeypatch.setenv("VLLM_ROCM_USE_AITER", "1")
if model == "TitanML/tiny-mixtral":
# Untrained model: near-uniform logits make argmax sensitive to
# AITER's bfloat16 rounding error in plain rms_norm.
# AITER's bfloat16 rounding error. Route the plain rms_norm and the
# fused MoE (whose near-uniform router logits flip expert selection
# under ~1 ULP drift) through the native kernels for this model.
# See ROCm/aiter#3806 for the tracking issue and minimal repro.
monkeypatch.setenv("VLLM_ROCM_USE_AITER_RMSNORM", "0")
monkeypatch.setenv("VLLM_ROCM_USE_AITER_MOE", "0")
elif use_rocm_aiter and model not in AITER_MODEL_LIST:
# Skip model that are not using AITER tests.
# When more AITER kernels are added, this list will not be