biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

Remove deprecated PyNcclConnector (#24151)

Browse Source 
Signed-off-by: Peter Pan <Peter.Pan@daocloud.io>
This commit is contained in:
Peter Pan
2025-09-04 06:49:16 +08:00
committed by GitHub
parent 36c260dad6
commit b5ee1e3261
7 changed files with 15 additions and 15 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
examples/offline_inference/disaggregated_prefill.py
View File

@@ -30,12 +30,12 @@ def run_prefill(prefill_done):
]
sampling_params = SamplingParams(temperature=0, top_p=0.95, max_tokens=1)
# Using PyNcclConnector to transmit KV caches between vLLM instances.
# Using P2pNcclConnector to transmit KV caches between vLLM instances.
# This instance is the prefill node (kv_producer, rank 0).
# The number of parallel instances for KV cache transfer is set to 2,
# as required for PyNcclConnector.
# as required for P2pNcclConnector.
ktc = KVTransferConfig(
kv_connector="PyNcclConnector",
kv_connector="P2pNcclConnector",
kv_role="kv_producer",
kv_rank=0,
kv_parallel_size=2,
@@ -74,12 +74,12 @@ def run_decode(prefill_done):
]
sampling_params = SamplingParams(temperature=0, top_p=0.95)
# Using PyNcclConnector to transmit KV caches between vLLM instances.
# Using P2pNcclConnector to transmit KV caches between vLLM instances.
# This instance is the decode node (kv_consumer, rank 1).
# The number of parallel instances for KV cache transfer is set to 2,
# as required for PyNcclConnector.
# as required for P2pNcclConnector.
ktc = KVTransferConfig(
kv_connector="PyNcclConnector",
kv_connector="P2pNcclConnector",
kv_role="kv_consumer",
kv_rank=1,
kv_parallel_size=2,

4
examples/online_serving/disaggregated_prefill.sh
View File

@@ -53,7 +53,7 @@ CUDA_VISIBLE_DEVICES=0 vllm serve $MODEL_NAME \
--gpu-memory-utilization 0.8 \
--trust-remote-code \
--kv-transfer-config \
'{"kv_connector":"PyNcclConnector","kv_role":"kv_producer","kv_rank":0,"kv_parallel_size":2}' &
'{"kv_connector":"P2pNcclConnector","kv_role":"kv_producer","kv_rank":0,"kv_parallel_size":2}' &
# decoding instance, which is the KV consumer
CUDA_VISIBLE_DEVICES=1 vllm serve $MODEL_NAME \
@@ -62,7 +62,7 @@ CUDA_VISIBLE_DEVICES=1 vllm serve $MODEL_NAME \
--gpu-memory-utilization 0.8 \
--trust-remote-code \
--kv-transfer-config \
'{"kv_connector":"PyNcclConnector","kv_role":"kv_consumer","kv_rank":1,"kv_parallel_size":2}' &
'{"kv_connector":"P2pNcclConnector","kv_role":"kv_consumer","kv_rank":1,"kv_parallel_size":2}' &
# wait until prefill and decode instances are ready
wait_for_server 8100