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 11, 00:00 UTC
[P/D] Mooncake connector support more protocols (#30133)
Signed-off-by: LCAIZJ <leichao139636@163.com>
Chao Lei committed
Dec 24, 2025 at 10:24 UTC
8b59753cdb5eb2a672fb44e3a281cebea197355b
1 file changed
+7
-1
vllm/distributed/kv_transfer/kv_connector/v1/mooncake_connector.py
+7
-1
@@ -408,7 +408,13 @@ class MooncakeConnectorWorker:
408
409
self.engine = TransferEngine()
410
self.hostname = get_ip()
411
- ret_value = self.engine.initialize(self.hostname, "P2PHANDSHAKE", "rdma", "")
411
+ protocol = self.vllm_config.kv_transfer_config.kv_connector_extra_config.get( # type: ignore[union-attr]
412
+ "mooncake_protocol", "rdma"
413
+ )
414
+ logger.info(
415
+ "The Mooncake Transfer Engine is using %s as its protocol.", protocol
416
+ )
417
+ ret_value = self.engine.initialize(self.hostname, "P2PHANDSHAKE", protocol, "")
418
if ret_value != 0:
419
raise RuntimeError("Mooncake Transfer Engine initialization failed.")
420