[ROCm][CI] Shard LM Eval Qwen3-5 Models (B200-MI355) in AMD CI (#46520)
Signed-off-by: Micah Williamson <micah.williamson@amd.com>
Micah Williamson committed
Jun 23, 2026 at 16:32 UTC
b28103e1ca8b697db917c9572bb0ba6e270c6c20
2 files changed
+5
-3
.buildkite/test-amd.yaml
+4
-3
@@ -2838,12 +2838,13 @@ steps:
2838
- uv pip install --system 'gpt-oss[eval]==0.0.5'
2839
- pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-gfx950.txt
2840
2841
-- label: LM Eval Qwen3-5 Models (B200-MI355) # TBD
2842
- timeout_in_minutes: 120
2841
+- label: LM Eval Qwen3-5 Models (B200-MI355) %N # TBD
2842
+ timeout_in_minutes: 180
2843
mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
2844
agent_pool: mi355_2
2845
num_gpus: 2
2846
optional: true
2847
+ parallelism: 4
2848
working_dir: "/vllm-workspace/tests"
2849
source_file_dependencies:
2850
- vllm/model_executor/models/qwen3_5.py
@@ -2858,7 +2859,7 @@ steps:
2859
- vllm/_aiter_ops.py
2860
- vllm/platforms/rocm.py
2861
commands:
2861
- - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-qwen35-mi355.txt
2862
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-qwen35-mi355.txt --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
2863
2864
- label: LM Eval Small Models (2xB200-2xMI355) # TBD
2865
timeout_in_minutes: 180
tests/evals/gsm8k/configs/Qwen3.5-35B-A3B-MXFP4-AITER-TP2.yaml
+1
@@ -3,6 +3,7 @@ accuracy_threshold: 0.89
3
tolerance: 0.03
4
num_questions: 1319
5
num_fewshot: 5
6
+startup_max_wait_seconds: 3600
7
server_args: >-
8
--max-model-len 4096
9
--tensor-parallel-size 2