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 14, 05:30 UTC
[Nixl][CI] Fix tests (#22806)
Signed-off-by: NickLucche <nlucches@redhat.com>
Nicolò Lucchesi committed
Aug 13, 2025 at 15:03 UTC
6b794c756c5a6b3c443c19a093435d02d91d525f
1 file changed
+3
tests/v1/kv_connector/unit/test_nixl_connector.py
+3
@@ -229,6 +229,9 @@ class FakeNixlConnectorWorker(NixlConnectorWorker):
229
num_blocks=1,
230
block_len=self.block_len,
231
attn_backend_name=self.backend_name,
232
+ # `self.kv_cache_layout` is only forced to HND when vllm engine
233
+ # is started. We mock HND here.
234
+ kv_cache_layout="HND",
235
),
236
remote_tp_size=remote_tp_size)
237
return {0: remote_agent_name}