[Frontend] Introduce Renderer for processing chat messages (using ModelConfig) (#30200)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
@@ -17,6 +17,7 @@ from vllm.multimodal.inputs import (
|
||||
MultiModalUUIDDict,
|
||||
)
|
||||
from vllm.multimodal.processing import BaseMultiModalProcessor
|
||||
from vllm.renderers import renderer_from_config
|
||||
from vllm.tokenizers import TokenizerLike
|
||||
from vllm.utils.jsontree import json_iter_leaves
|
||||
from vllm.v1.metrics.stats import MultiModalCacheStats
|
||||
@@ -46,7 +47,6 @@ class InputPreprocessor:
|
||||
def __init__(
|
||||
self,
|
||||
model_config: ModelConfig,
|
||||
tokenizer: TokenizerLike | None,
|
||||
observability_config: ObservabilityConfig | None = None,
|
||||
mm_registry: MultiModalRegistry = MULTIMODAL_REGISTRY,
|
||||
mm_processor_cache: BaseMultiModalProcessorCache | None = None,
|
||||
@@ -54,20 +54,19 @@ class InputPreprocessor:
|
||||
super().__init__()
|
||||
|
||||
self.model_config = model_config
|
||||
self.tokenizer = tokenizer
|
||||
self.observability_config = observability_config
|
||||
self.renderer = renderer_from_config(model_config)
|
||||
self.mm_registry = mm_registry
|
||||
self.mm_processor_cache = mm_processor_cache
|
||||
|
||||
self.mm_cache_stats = MultiModalCacheStats() if mm_processor_cache else None
|
||||
|
||||
def get_tokenizer(self) -> TokenizerLike:
|
||||
if self.tokenizer is None:
|
||||
raise ValueError(
|
||||
"You cannot pass text prompts when `skip_tokenizer_init=True`"
|
||||
)
|
||||
@property
|
||||
def tokenizer(self) -> TokenizerLike | None:
|
||||
return self.renderer.tokenizer
|
||||
|
||||
return self.tokenizer
|
||||
def get_tokenizer(self) -> TokenizerLike:
|
||||
return self.renderer.get_tokenizer()
|
||||
|
||||
def get_bos_token_id(self) -> int | None:
|
||||
if self.tokenizer is None:
|
||||
|
||||
Reference in New Issue
Block a user