A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 10, 19:00 UTC
[CI] Skip Phi-MoE test due to old API util (#31632)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Jan 4, 2026 at 18:52 UTC
89f1f253109961aea9c09062e66ae28134fa4746
2 files changed
+15
-2
.buildkite/test-amd.yaml
+2
-2
@@ -889,7 +889,7 @@ steps:
889
commands:
890
# Install fast path packages for testing against transformers
891
# Note: also needed to run plamo2 model in vLLM
892
- - uv pip install --system --no-build-isolation 'git+https://github.com/state-spaces/mamba@v2.2.5'
892
+ - uv pip install --system --no-build-isolation 'git+https://github.com/AndreasKaratzas/mamba@fix-rocm-7.0-warp-size-constexpr'
893
- uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.2'
894
# Shard hybrid language model tests
895
- pytest -v -s models/language/generation \
@@ -910,7 +910,7 @@ steps:
910
commands:
911
# Install fast path packages for testing against transformers
912
# Note: also needed to run plamo2 model in vLLM
913
- - uv pip install --system --no-build-isolation 'git+https://github.com/state-spaces/mamba@v2.2.5'
913
+ - uv pip install --system --no-build-isolation 'git+https://github.com/AndreasKaratzas/mamba@fix-rocm-7.0-warp-size-constexpr'
914
- uv pip install --system --no-build-isolation 'git+https://github.com/Dao-AILab/causal-conv1d@v1.5.2'
915
- pytest -v -s models/language/generation -m '(not core_model) and (not hybrid_model)'
916
tests/models/language/generation/test_phimoe.py
+13
@@ -60,6 +60,19 @@ def test_phimoe_routing_function():
60
assert torch.equal(topk_ids, ground_truth[test_id]["topk_ids"])
61
62
63
+# There is a known issue that triggers `AttributeError: 'DynamicCache'
64
+# object has no attribute 'seen_tokens'` when running:
65
+# `tests/models/language/generation/test_phimoe.py::test_models
66
+# [5-64-bfloat16-microsoft/Phi-3.5-MoE-instruct]`
67
+# This issue is being investigated and tracked in:
68
+# https://huggingface.co/microsoft/Phi-3.5-MoE-instruct/discussions/58
69
+# It is platform-agnostic. Therefore, we skip this test on all platforms for now.
70
+@pytest.mark.skip(
71
+ reason="Skipping due to known issue: "
72
+ "'DynamicCache' object has no attribute 'seen_tokens'. See: "
73
+ "https://huggingface.co/microsoft/Phi-3.5-MoE-instruct/discussions/58 "
74
+ "for details.",
75
+)
76
@pytest.mark.skipif(
77
condition=current_platform.is_cpu(),
78
reason="This test takes a lot time to run on CPU, "