iHore/vLLM
8
0
Fork 0
Code Issues Pull Requests Actions 1 Packages Projects Releases Wiki Activity

[Bugfix][KVConnector] Support DCP/PCP in OffloadingConnector (#41549)

Browse Source 
Signed-off-by: Itay Etelis <itay.etelis@ibm.com>
Co-authored-by: Itay Etelis <itay.etelis@ibm.com>
This commit is contained in:
Itay Etelis
2026-05-05 14:54:29 +03:00
committed by GitHub
co-authored by Itay Etelis
parent b0765bee17
commit 98661fe012
1 changed files with 10 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files
vllm/v1/kv_offload/base.py
+10 -2
View File
@@ -331,12 +331,20 @@ class OffloadingSpec(ABC):
assert kv_transfer_config is not None
self.extra_config = kv_transfer_config.kv_connector_extra_config
parallel_config = vllm_config.parallel_config
context_parallel_factor = (
parallel_config.decode_context_parallel_size
* parallel_config.prefill_context_parallel_size
)
# block size used by vLLM for hashing request tokens for the sake
# of enabling prefix caching
self.hash_block_size = vllm_config.cache_config.block_size
self.hash_block_size = (
vllm_config.cache_config.block_size * context_parallel_factor
)
# gpu block size per group
self.gpu_block_size: tuple[int, ...] = tuple(
kv_cache_group.kv_cache_spec.block_size
kv_cache_group.kv_cache_spec.block_size * context_parallel_factor
for kv_cache_group in kv_cache_config.kv_cache_groups
)