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 6, 02:00 UTC
[ROCm][CI][PD] Add Hybrid SSM integration tests to CI (#37924)
Signed-off-by: Andreas Karatzas <akaratza@amd.com>
Andreas Karatzas committed
Mar 24, 2026 at 18:58 UTC
8bbb7c7f200ea791eb4ccd15e77433e29e13ec95
1 file changed
+15
.buildkite/test-amd.yaml
+15
@@ -1387,6 +1387,21 @@ steps:
1387
- CROSS_LAYERS_BLOCKS=True ROCM_ATTN=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
1388
1389
1390
+- label: Hyrbid SSM NixlConnector PD accuracy tests (4 GPUs) # TBD
1391
+ timeout_in_minutes: 180
1392
+ mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]
1393
+ agent_pool: mi250_4
1394
+ num_gpus: 4
1395
+ working_dir: "/vllm-workspace/tests"
1396
+ source_file_dependencies:
1397
+ - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
1398
+ - tests/v1/kv_connector/nixl_integration/
1399
+ - vllm/platforms/rocm.py
1400
+ commands:
1401
+ - uv pip install --system -r /vllm-workspace/requirements/kv_connectors_rocm.txt
1402
+ - HYBRID_SSM=1 ROCM_ATTN=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
1403
+
1404
+
1405
- label: Distributed Tests (2 GPUs)(H100-MI250) # TBD
1406
timeout_in_minutes: 180
1407
mirror_hardwares: [amdexperimental, amdproduction, amdgfx90anightly, amdmi250]