[ROCm][CI] Skip Qwen3.5-35B-A3B-MXFP4-AITER-TP2 for non gfx950 (#46109)
Signed-off-by: charlifu <charlifu@amd.com>
Charlie Fu committed
Jun 19, 2026 at 17:20 UTC
e6cd8913ddfe63b4620e45ff8c2da1d37318dbe5
2 files changed
+18
.buildkite/test_areas/lm_eval.yaml
+9
@@ -101,6 +101,15 @@ steps:
101
num_devices: 8
102
commands:
103
- pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-h200.txt
104
+ mirror:
105
+ amd:
106
+ device: mi300_8
107
+ timeout_in_minutes: 180
108
+ depends_on:
109
+ - image-build-amd
110
+ commands:
111
+ - export VLLM_WORKER_MULTIPROC_METHOD=spawn
112
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-mi3xx.txt
113
114
- label: MoE Refactor Integration Test (H100 - TEMPORARY)
115
key: moe-refactor-integration-test-h100-temporary
tests/evals/gsm8k/test_gsm8k_correctness.py
+9
@@ -78,7 +78,16 @@ def test_gsm8k_correctness(config_filename):
78
"Skipping DeepSeek-V3.2 and DeepSeek-R1 on ROCm platforms "
79
"due to agent pool disk space issues and pod evictions."
80
)
81
+ if current_platform.is_rocm() and (
82
+ "Qwen3.5-35B-A3B-MXFP4-AITER-TP2" in config_filename.name
83
+ ):
84
+ from vllm.platforms.rocm import on_gfx950
85
86
+ if not on_gfx950():
87
+ pytest.skip(
88
+ "Skipping Qwen3.5-35B-A3B-MXFP4-AITER-TP2 on non-GFX950 platforms. "
89
+ "The quantization scheme is not supported on non-GFX950 platforms."
90
+ )
91
# Parse server arguments from config (use shlex to handle quoted strings)
92
server_args_str = eval_config.get("server_args", "")
93
server_args = shlex.split(server_args_str) if server_args_str else []