frozenleaves/vllm
1
0
Fork 0
mirror of https://github.com/vllm-project/vllm.git synced 2025-10-20 14:53:52 +08:00
Code Issues Packages Projects Releases Wiki Activity

skip fusedmoe layer for start_load_kv (#21378)

Browse Source 
Signed-off-by: calvin chen <wen.chen@dynamia.ai>
This commit is contained in:
Calvin Chen
2025-07-29 09:59:44 +08:00
committed by GitHub
parent afa2607596
commit e18f085103
1 changed files with 10 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/distributed/kv_transfer/kv_connector/v1/p2p/p2p_nccl_connector.py
View File

@ -192,8 +192,16 @@ class P2pNcclConnector(KVConnectorBase_V1):
# Load the KV for each request each layer
for request in metadata.requests:
for layer_name in forward_context.no_compile_layers:
attn_layer = forward_context.no_compile_layers[layer_name]
kv_cache_layer = attn_layer.kv_cache[ \
layer = forward_context.no_compile_layers[layer_name]
# Only process layers that have kv_cache
# attribute (attention layers) Skip non-attention
# layers like FusedMoE
kv_cache = getattr(layer, 'kv_cache', None)
if kv_cache is None:
continue
kv_cache_layer = kv_cache[ \
forward_context.virtual_engine]
kv_cache = self.p2p_nccl_engine.recv_tensor(