From 3fd1d4ec2c992689594f9c7ee0ac79597f74a2ef Mon Sep 17 00:00:00 2001 From: Charlie Fu Date: Mon, 2 Mar 2026 01:43:38 -0600 Subject: [PATCH] [Rocm][CI] Fix LM Eval Large Models (H100) test group (#34750) Signed-off-by: charlifu --- .buildkite/lm-eval-harness/configs/models-large-rocm.txt | 1 + .buildkite/test-amd.yaml | 8 ++++---- 2 files changed, 5 insertions(+), 4 deletions(-) diff --git a/.buildkite/lm-eval-harness/configs/models-large-rocm.txt b/.buildkite/lm-eval-harness/configs/models-large-rocm.txt index 4fb0b84bc4..a9a60f348d 100644 --- a/.buildkite/lm-eval-harness/configs/models-large-rocm.txt +++ b/.buildkite/lm-eval-harness/configs/models-large-rocm.txt @@ -1 +1,2 @@ Meta-Llama-4-Maverick-17B-128E-Instruct-FP8.yaml +Qwen3-235B-A22B-Instruct-2507-FP8.yaml diff --git a/.buildkite/test-amd.yaml b/.buildkite/test-amd.yaml index 6c35e0db13..ab8bf9d23c 100644 --- a/.buildkite/test-amd.yaml +++ b/.buildkite/test-amd.yaml @@ -1544,8 +1544,8 @@ steps: - export VLLM_WORKER_MULTIPROC_METHOD=spawn - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large.txt --tp-size=4 -##### H100 test ##### -- label: LM Eval Large Models (H100) # optional +##### FP8 test ##### +- label: LM Eval Large Models (H100) # optional, still use H100 for consistency gpu: h100 optional: true mirror_hardwares: [amdexperimental, amdproduction] @@ -1557,8 +1557,8 @@ steps: - csrc/ - vllm/model_executor/layers/quantization commands: - - export VLLM_USE_DEEP_GEMM=0 # We found Triton is faster than DeepGEMM for H100 - - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-hopper.txt --tp-size=4 + - export VLLM_USE_DEEP_GEMM=0 + - pytest -s -v test_lm_eval_correctness.py --config-list-file=configs/models-large-rocm.txt --tp-size=4 ##### H200 test #####