A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 10, 04:00 UTC
[code clean] remove duplicate check (#32376)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Jan 15, 2026 at 13:29 UTC
9d7ae3fcdbdecf828c5189f0152808fcbe6dbc86
3 files changed
+3
-9
vllm/multimodal/registry.py
+1
-1
@@ -254,7 +254,7 @@ class MultiModalRegistry:
254
observability_config: "ObservabilityConfig | None" = None,
255
tokenizer: TokenizerLike | None = None,
256
) -> InputProcessingContext:
257
- if tokenizer is None and not model_config.skip_tokenizer_init:
257
+ if tokenizer is None:
258
tokenizer = cached_tokenizer_from_config(model_config)
259
260
return InputProcessingContext(
vllm/v1/engine/async_llm.py
+1
-4
@@ -106,10 +106,7 @@ class AsyncLLM(EngineClient):
106
"enabling logging without default stat loggers."
107
)
108
109
- if self.model_config.skip_tokenizer_init:
110
- tokenizer = None
111
- else:
112
- tokenizer = cached_tokenizer_from_config(self.model_config)
109
+ tokenizer = cached_tokenizer_from_config(self.model_config)
110
111
self.input_processor = InputProcessor(self.vllm_config, tokenizer)
112
self.io_processor = get_io_processor(
vllm/v1/engine/llm_engine.py
+1
-4
@@ -84,10 +84,7 @@ class LLMEngine:
84
self.dp_group = None
85
self.should_execute_dummy_batch = False
86
87
- if self.model_config.skip_tokenizer_init:
88
- tokenizer = None
89
- else:
90
- tokenizer = cached_tokenizer_from_config(self.model_config)
87
+ tokenizer = cached_tokenizer_from_config(self.model_config)
88
89
self.input_processor = InputProcessor(self.vllm_config, tokenizer)
90
self.io_processor = get_io_processor(