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 8, 05:00 UTC
[CI][Nixl] Add CrossLayer KV layout tests (#34615)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Feb 17, 2026 at 14:35 UTC
8e962fef5fec5197d8c027ed2e00ecace5b47aac
1 file changed
+11
.buildkite/test_areas/distributed.yaml
+11
@@ -197,6 +197,17 @@ steps:
197
- uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
198
- DP_EP=1 bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
199
200
+- label: CrossLayer KV layout Distributed NixlConnector PD accuracy tests (4 GPUs)
201
+ timeout_in_minutes: 30
202
+ working_dir: "/vllm-workspace/tests"
203
+ num_devices: 4
204
+ source_file_dependencies:
205
+ - vllm/distributed/kv_transfer/kv_connector/v1/nixl_connector.py
206
+ - tests/v1/kv_connector/nixl_integration/
207
+ commands:
208
+ - uv pip install --system -r /vllm-workspace/requirements/kv_connectors.txt
209
+ - CROSS_LAYERS_BLOCKS=True bash v1/kv_connector/nixl_integration/config_sweep_accuracy_test.sh
210
+
211
- label: Pipeline + Context Parallelism (4 GPUs))
212
timeout_in_minutes: 60
213
working_dir: "/vllm-workspace/tests"