xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-15 06:45:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[KV Connector] Update lmcache connector with latest compatibility (#27681)

Browse Source 
Signed-off-by: Samuel Shen <slshen@uchicago.edu>
Co-authored-by: Samuel Shen <slshen@uchicago.edu>
This commit is contained in:
Shaoting 2025-10-28 22:38:37 -07:00 committed by GitHub
parent 0d8161b075
commit a4a4f0f617
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 20 additions and 2 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

22
vllm/distributed/kv_transfer/kv_connector/v1/lmcache_integration/vllm_v1_adapter.py
View File

@ -44,8 +44,8 @@ from vllm.distributed.kv_transfer.kv_connector.v1.lmcache_integration.utils impo
) )
from vllm.distributed.parallel_state import get_tensor_model_parallel_rank, get_tp_group from vllm.distributed.parallel_state import get_tensor_model_parallel_rank, get_tp_group
from vllm.sampling_params import SamplingParams from vllm.sampling_params import SamplingParams
from vllm.utils import get_kv_cache_torch_dtype
from vllm.utils.math_utils import cdiv from vllm.utils.math_utils import cdiv
from vllm.utils.torch_utils import get_kv_cache_torch_dtype
from vllm.v1.core.sched.output import SchedulerOutput from vllm.v1.core.sched.output import SchedulerOutput
from vllm.version import __version__ as VLLM_VERSION from vllm.version import __version__ as VLLM_VERSION
@ -389,7 +389,7 @@ class ReqMeta:
def need_gpu_interm_buffer(lmcache_config: LMCacheEngineConfig): def need_gpu_interm_buffer(lmcache_config: LMCacheEngineConfig):
return lmcache_config.enable_pd return not lmcache_config.enable_pd
def _calculate_mtp_layers(vllm_config, model_config): def _calculate_mtp_layers(vllm_config, model_config):
@ -403,6 +403,20 @@ def _calculate_mtp_layers(vllm_config, model_config):
num_mtp_layers = getattr( num_mtp_layers = getattr(
model_config.hf_config, "num_nextn_predict_layers", 0 model_config.hf_config, "num_nextn_predict_layers", 0
) )
elif vllm_config.speculative_config.use_eagle():
try:
draft_model_config = vllm_config.speculative_config.draft_model_config
num_mtp_layers = draft_model_config.get_num_layers(
vllm_config.parallel_config
)
logger.info("EAGLE detected %d extra layer(s)", num_mtp_layers)
except Exception:
logger.info(
"EAGLE detected, but failed to get the number of extra layers"
"falling back to 1"
)
num_mtp_layers = 1
return num_mtp_layers return num_mtp_layers
@ -1208,6 +1222,10 @@ class LMCacheConnectorV1Impl:
if the CacheManager this allocated blocks for us. if the CacheManager this allocated blocks for us.
""" """
# Clear local status in lookup client when a new request is
# successfully scheduled.
self.lookup_client.clear_lookup_status(request.request_id)
kv_transfer_params = ( kv_transfer_params = (
request.kv_transfer_params request.kv_transfer_params
if hasattr(request, "kv_transfer_params") if hasattr(request, "kv_transfer_params")