From 9c3ae04bfe6532d6d6d812ae6688366c9911555b Mon Sep 17 00:00:00 2001 From: Andreas Karatzas Date: Thu, 26 Mar 2026 11:51:18 -0500 Subject: [PATCH] [ROCm][CI] Add LM Eval Qwen3.5 Models test for MI355 (#38155) Signed-off-by: Andreas Karatzas --- .buildkite/test-amd.yaml | 24 +++++++++++++++++++ .../gsm8k/configs/models-qwen35-mi355.txt | 1 + 2 files changed, 25 insertions(+) create mode 100644 tests/evals/gsm8k/configs/models-qwen35-mi355.txt diff --git a/.buildkite/test-amd.yaml b/.buildkite/test-amd.yaml index c0cb18bd8..4b2494416 100644 --- a/.buildkite/test-amd.yaml +++ b/.buildkite/test-amd.yaml @@ -3749,3 +3749,27 @@ steps: - vllm/platforms/rocm.py commands: - python3 benchmarks/attention_benchmarks/benchmark.py --backends ROCM_ATTN ROCM_AITER_FA ROCM_AITER_UNIFIED_ATTN --batch-specs "8q1s1k" --repeats 1 --warmup-iters 1 + + +- label: LM Eval Qwen3-5 Models (B200-MI355) # TBD + timeout_in_minutes: 120 + mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355] + agent_pool: mi355_2 + num_gpus: 2 + optional: true + working_dir: "/vllm-workspace/tests" + source_file_dependencies: + - vllm/model_executor/models/qwen3_5.py + - vllm/model_executor/models/qwen3_5_mtp.py + - vllm/transformers_utils/configs/qwen3_5.py + - vllm/transformers_utils/configs/qwen3_5_moe.py + - vllm/model_executor/models/qwen.py + - vllm/model_executor/models/qwen2.py + - vllm/model_executor/models/qwen3.py + - vllm/model_executor/models/qwen3_next.py + - vllm/model_executor/models/qwen3_next_mtp.py + - vllm/model_executor/layers/fla/ops/ + - vllm/_aiter_ops.py + - vllm/platforms/rocm.py + commands: + - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-qwen35-mi355.txt diff --git a/tests/evals/gsm8k/configs/models-qwen35-mi355.txt b/tests/evals/gsm8k/configs/models-qwen35-mi355.txt new file mode 100644 index 000000000..4e7af71c7 --- /dev/null +++ b/tests/evals/gsm8k/configs/models-qwen35-mi355.txt @@ -0,0 +1 @@ +Qwen3.5-35B-A3B-DEP2.yaml