From 82a006beebf03c4f7bd600ab68f15b3325feb8e4 Mon Sep 17 00:00:00 2001 From: Bowen Bao Date: Thu, 2 Apr 2026 09:06:01 -0700 Subject: [PATCH] [CI][ROCm] Add gpt-oss w4a8 in CI (#38292) Signed-off-by: Bowen Bao --- .../evals/gpt_oss/configs/gpt-oss-20b-rocm-mxfp4-fp8.yaml | 8 ++++++++ tests/evals/gpt_oss/configs/models-gfx950.txt | 3 ++- 2 files changed, 10 insertions(+), 1 deletion(-) create mode 100644 tests/evals/gpt_oss/configs/gpt-oss-20b-rocm-mxfp4-fp8.yaml diff --git a/tests/evals/gpt_oss/configs/gpt-oss-20b-rocm-mxfp4-fp8.yaml b/tests/evals/gpt_oss/configs/gpt-oss-20b-rocm-mxfp4-fp8.yaml new file mode 100644 index 0000000000..f7dd14784a --- /dev/null +++ b/tests/evals/gpt_oss/configs/gpt-oss-20b-rocm-mxfp4-fp8.yaml @@ -0,0 +1,8 @@ +# SPDX-License-Identifier: Apache-2.0 +# SPDX-FileCopyrightText: Copyright contributors to the vLLM project +model_name: amd/gpt-oss-20b-MoE-Quant-W-MXFP4-A-FP8-KV-FP8 +metric_threshold: 0.568 +reasoning_effort: low +server_args: "--attention-backend ROCM_AITER_UNIFIED_ATTN" +env: + VLLM_ROCM_USE_AITER: "1" \ No newline at end of file diff --git a/tests/evals/gpt_oss/configs/models-gfx950.txt b/tests/evals/gpt_oss/configs/models-gfx950.txt index 2b6ff4f4a8..5085aa9f2e 100644 --- a/tests/evals/gpt_oss/configs/models-gfx950.txt +++ b/tests/evals/gpt_oss/configs/models-gfx950.txt @@ -1,3 +1,4 @@ # GFX950 model configurations for GPQA evaluation # Tests different environment variable combinations -gpt-oss-20b-rocm-baseline.yaml \ No newline at end of file +gpt-oss-20b-rocm-baseline.yaml +gpt-oss-20b-rocm-mxfp4-fp8.yaml \ No newline at end of file