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 5, 23:30 UTC
[ROCm][CI] Add LM Eval Qwen3.5 Models test for MI355 (#38155)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 26, 2026 at 11:51 UTC
9c3ae04bfe6532d6d6d812ae6688366c9911555b
2 files changed
+25
.buildkite/test-amd.yaml
+24
@@ -3749,3 +3749,27 @@ steps:
3749
- vllm/platforms/rocm.py
3750
commands:
3751
- python3 benchmarks/attention_benchmarks/benchmark.py --backends ROCM_ATTN ROCM_AITER_FA ROCM_AITER_UNIFIED_ATTN --batch-specs "8q1s1k" --repeats 1 --warmup-iters 1
3752
+
3753
+
3754
+- label: LM Eval Qwen3-5 Models (B200-MI355) # TBD
3755
+ timeout_in_minutes: 120
3756
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx950nightly, amdmi355]
3757
+ agent_pool: mi355_2
3758
+ num_gpus: 2
3759
+ optional: true
3760
+ working_dir: "/vllm-workspace/tests"
3761
+ source_file_dependencies:
3762
+ - vllm/model_executor/models/qwen3_5.py
3763
+ - vllm/model_executor/models/qwen3_5_mtp.py
3764
+ - vllm/transformers_utils/configs/qwen3_5.py
3765
+ - vllm/transformers_utils/configs/qwen3_5_moe.py
3766
+ - vllm/model_executor/models/qwen.py
3767
+ - vllm/model_executor/models/qwen2.py
3768
+ - vllm/model_executor/models/qwen3.py
3769
+ - vllm/model_executor/models/qwen3_next.py
3770
+ - vllm/model_executor/models/qwen3_next_mtp.py
3771
+ - vllm/model_executor/layers/fla/ops/
3772
+ - vllm/_aiter_ops.py
3773
+ - vllm/platforms/rocm.py
3774
+ commands:
3775
+ - pytest -s -v evals/gsm8k/test_gsm8k_correctness.py --config-list-file=configs/models-qwen35-mi355.txt
tests/evals/gsm8k/configs/models-qwen35-mi355.txt
new
+1
@@ -0,0 +1 @@
1
+Qwen3.5-35B-A3B-DEP2.yaml