[Bugfix] Fix SHM cache initialization (#26427)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-10-09 02:48:04 -07:00
committed by GitHub
parent dc7976dd9f
commit 4bdf7ac593
30 changed files with 357 additions and 417 deletions
+20 -17
View File
@@ -19,11 +19,12 @@ from vllm.logger import init_logger
from vllm.lora.request import LoRARequest
from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
from vllm.outputs import PoolingRequestOutput, RequestOutput
from vllm.plugins.io_processors import get_io_processor
from vllm.pooling_params import PoolingParams
from vllm.sampling_params import SamplingParams
from vllm.tasks import SupportedTask
from vllm.tracing import init_tracer
from vllm.transformers_utils.tokenizer import AnyTokenizer
from vllm.transformers_utils.tokenizer import AnyTokenizer, init_tokenizer_from_configs
from vllm.usage.usage_lib import UsageContext
from vllm.utils import Device
from vllm.v1.engine import EngineCoreRequest
@@ -95,8 +96,16 @@ class LLMEngine:
self.dp_group = None
self.should_execute_dummy_batch = False
# Processor (convert Inputs --> EngineCoreRequests)
self.processor = Processor(vllm_config, mm_registry=mm_registry)
if self.model_config.skip_tokenizer_init:
tokenizer = None
else:
tokenizer = init_tokenizer_from_configs(self.model_config)
self.processor = Processor(self.vllm_config, tokenizer)
self.io_processor = get_io_processor(
self.vllm_config,
self.model_config.io_processor_plugin,
)
# OutputProcessor (convert EngineCoreOutputs --> RequestOutput).
self.output_processor = OutputProcessor(
@@ -204,14 +213,6 @@ class LLMEngine:
def validate_outputs(cls, outputs, output_type):
return outputs
@property
def tokenizer(self) -> Optional[AnyTokenizer]:
return self.processor.tokenizer
@tokenizer.setter
def tokenizer(self, tokenizer: Optional[AnyTokenizer]) -> None:
self.processor.tokenizer = tokenizer
def get_supported_tasks(self) -> tuple[SupportedTask, ...]:
return self.engine_core.get_supported_tasks()
@@ -313,12 +314,6 @@ class LLMEngine:
return processed_outputs.request_outputs
def get_vllm_config(self):
return self.vllm_config
def get_model_config(self):
return self.model_config
def start_profile(self):
self.engine_core.profile(True)
@@ -345,6 +340,14 @@ class LLMEngine:
assert self.log_stats, "Stat logging disabled"
return get_metrics_snapshot()
@property
def tokenizer(self) -> Optional[AnyTokenizer]:
return self.processor.tokenizer
@tokenizer.setter
def tokenizer(self, tokenizer: Optional[AnyTokenizer]) -> None:
self.processor.tokenizer = tokenizer
def get_tokenizer(self) -> AnyTokenizer:
if self.tokenizer is None:
raise ValueError(