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

[KVConnector] Clean up redundant code in KV connectors (#34147)

Browse Source 
Signed-off-by: Martin Hickey <martin.hickey@ie.ibm.com>
This commit is contained in:
Martin Hickey
2026-02-13 08:14:30 +00:00
committed by GitHub
parent 934acddef9
commit 47e9b63e1a
4 changed files with 2 additions and 11 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/distributed/kv_transfer/kv_connector/v1/__init__.py
View File

@@ -6,7 +6,7 @@ from vllm.distributed.kv_transfer.kv_connector.v1.base import (
SupportsHMA, SupportsHMA,
supports_hma, supports_hma,
) )
from vllm.distributed.kv_transfer.kv_connector.v1.decode_bench_connector import ( # noqa E:501 from vllm.distributed.kv_transfer.kv_connector.v1.decode_bench_connector import ( # noqa: E501
DecodeBenchConnector, DecodeBenchConnector,
) )

8
vllm/distributed/kv_transfer/kv_connector/v1/example_connector.py
View File

@@ -145,7 +145,6 @@ class ExampleConnector(KVConnectorBase_V1):
num_pages * page_size, -1 num_pages * page_size, -1
) )
dst_kv_cache_layer[slot_mapping, ...] = src_kv_cache dst_kv_cache_layer[slot_mapping, ...] = src_kv_cache
dst_kv_cache_layer.reshape(dst_kv_cache_layer_shape)
else: else:
num_pages = dst_kv_cache_layer_shape[1] num_pages = dst_kv_cache_layer_shape[1]
page_size = dst_kv_cache_layer_shape[2] page_size = dst_kv_cache_layer_shape[2]
@@ -153,18 +152,11 @@ class ExampleConnector(KVConnectorBase_V1):
2, num_pages * page_size, -1 2, num_pages * page_size, -1
) )
dst_kv_cache_layer[:, slot_mapping, ...] = src_kv_cache dst_kv_cache_layer[:, slot_mapping, ...] = src_kv_cache
dst_kv_cache_layer.reshape(dst_kv_cache_layer_shape)
# Get the metadata # Get the metadata
metadata: KVConnectorMetadata = self._get_connector_metadata() metadata: KVConnectorMetadata = self._get_connector_metadata()
assert isinstance(metadata, ExampleConnectorMetadata) assert isinstance(metadata, ExampleConnectorMetadata)
if metadata is None:
logger.warning(
"In connector.start_load_kv, but the connector metadata is None"
)
return
attn_metadata = forward_context.attn_metadata attn_metadata = forward_context.attn_metadata
if attn_metadata is None: if attn_metadata is None:
logger.warning("In connector.start_load_kv, but the attn_metadata is None") logger.warning("In connector.start_load_kv, but the attn_metadata is None")

1
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_mp_connector.py
View File

@@ -36,7 +36,6 @@ except ImportError:
) )
if TYPE_CHECKING: if TYPE_CHECKING:
from vllm.config import VllmConfig
from vllm.distributed.kv_events import KVCacheEvent from vllm.distributed.kv_events import KVCacheEvent
from vllm.distributed.kv_transfer.kv_connector.v1.metrics import ( from vllm.distributed.kv_transfer.kv_connector.v1.metrics import (
KVConnectorPromMetrics, KVConnectorPromMetrics,

2
vllm/distributed/kv_transfer/kv_connector/v1/offloading_connector.py
View File

@@ -85,7 +85,7 @@ class OffloadingConnectorStats(KVConnectorStats):
for transfer_type, ops_list in self.data.items(): for transfer_type, ops_list in self.data.items():
assert isinstance(ops_list, list) assert isinstance(ops_list, list)
total_bytes = 0 total_bytes = 0
total_time = 0 total_time = 0.0
for op in ops_list: for op in ops_list:
assert isinstance(op, dict) assert isinstance(op, dict)
total_bytes += op["op_size"] total_bytes += op["op_size"]