| 1 | # SPDX-License-Identifier: Apache-2.0 |
| 2 | # SPDX-FileCopyrightText: Copyright contributors to the vLLM project |
| 3 | model_name: amd/gpt-oss-20b-MoE-Quant-W-MXFP4-A-FP8-KV-FP8 |
| 4 | metric_threshold: 0.568 |
| 5 | reasoning_effort: low |
| 6 | server_args: "--attention-backend ROCM_AITER_UNIFIED_ATTN --tensor-parallel-size 2" |
| 7 | env: |
| 8 | VLLM_ROCM_USE_AITER: "1" |