A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 7, 00:15 UTC
[ROCm][CI] Fix ROCm GPT-OSS Eval test group (#36179)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 9, 2026 at 19:55 UTC
179547d62c73e7174bf42b8ca0a34177ac3a5c9e
4 files changed
+16
-4
.buildkite/test-amd.yaml
+4
-4
@@ -1639,8 +1639,8 @@ steps:
1639
- vllm/model_executor/layers/quantization/mxfp4.py
1640
- vllm/v1/attention/backends/flashinfer.py
1641
commands:
1642
- - uv pip install --system 'gpt-oss[eval]==0.0.5'
1643
- - VLLM_ROCM_USE_AITER_MHA=0 VLLM_ROCM_USE_AITER=1 VLLM_USE_AITER_UNIFIED_ATTENTION=1 pytest -s -v tests/evals/gpt_oss/test_gpqa_correctness.py --model openai/gpt-oss-20b --metric 0.58
1642
+ - uv pip install --system 'gpt-oss[eval]==0.0.5'
1643
+ - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-gfx942.txt
1644
1645
##### EPLB Accuracy Tests #####
1646
- label: DeepSeek V2-Lite Accuracy
@@ -3296,8 +3296,8 @@ steps:
3296
- vllm/model_executor/layers/quantization/mxfp4.py
3297
- vllm/v1/attention/backends/flashinfer.py
3298
commands:
3299
- - uv pip install --system 'gpt-oss[eval]==0.0.5'
3300
- - VLLM_ROCM_USE_AITER_MHA=0 VLLM_ROCM_USE_AITER=1 VLLM_USE_AITER_UNIFIED_ATTENTION=1 pytest -s -v tests/evals/gpt_oss/test_gpqa_correctness.py --model openai/gpt-oss-20b --metric 0.58
3299
+ - uv pip install --system 'gpt-oss[eval]==0.0.5'
3300
+ - pytest -s -v evals/gpt_oss/test_gpqa_correctness.py --config-list-file=configs/models-gfx950.txt
3301
3302
##### EPLB Accuracy Tests #####
3303
- label: DeepSeek V2-Lite Accuracy
tests/evals/gpt_oss/configs/gpt-oss-20b-rocm-baseline.yaml
new
+6
@@ -0,0 +1,6 @@
1
+# SPDX-License-Identifier: Apache-2.0
2
+# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
+model_name: openai/gpt-oss-20b
4
+metric_threshold: 0.568
5
+reasoning_effort: low
6
+server_args: "--attention-backend ROCM_AITER_UNIFIED_ATTN"
\ No newline at end of file
tests/evals/gpt_oss/configs/models-gfx942.txt
new
+3
@@ -0,0 +1,3 @@
1
+# GFX942 model configurations for GPQA evaluation
2
+# Tests different environment variable combinations
3
+gpt-oss-20b-rocm-baseline.yaml
\ No newline at end of file
tests/evals/gpt_oss/configs/models-gfx950.txt
new
+3
@@ -0,0 +1,3 @@
1
+# GFX950 model configurations for GPQA evaluation
2
+# Tests different environment variable combinations
3
+gpt-oss-20b-rocm-baseline.yaml
\ No newline at end of file