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 16, 07:00 UTC
[Misc] Suppress duplicated logging regarding multimodal input pipeline (#10530)
Signed-off-by: Roger Wang <ywang@roblox.com>
Roger Wang committed
Nov 21, 2024 at 09:21 UTC
c51e397fe8db2ef0664814ef3f80e1237c7283da
2 files changed
+8
-2
vllm/inputs/preprocess.py
+2
-2
@@ -10,7 +10,7 @@ from vllm.multimodal import MULTIMODAL_REGISTRY, MultiModalRegistry
10
from vllm.multimodal.processing import MultiModalDataDict, MultiModalInputsV2
11
from vllm.prompt_adapter.request import PromptAdapterRequest
12
from vllm.transformers_utils.tokenizer_group import BaseTokenizerGroup
13
-from vllm.utils import print_warning_once
13
+from vllm.utils import print_info_once, print_warning_once
14
15
from .data import (DecoderOnlyInputs, EncoderDecoderInputs, ProcessorInputs,
16
PromptType, SingletonInputs, SingletonPrompt, token_inputs)
@@ -212,7 +212,7 @@ class InputPreprocessor:
212
# updated to use the new multi-modal processor
213
can_process_multimodal = self.mm_registry.has_processor(model_config)
214
if not can_process_multimodal:
215
- logger.info(
215
+ print_info_once(
216
"Your model uses the legacy input pipeline instead of the new "
217
"multi-modal processor. Please note that the legacy pipeline "
218
"will be removed in a future release. For more details, see: "
vllm/utils.py
+6
@@ -705,6 +705,12 @@ def create_kv_caches_with_random(
705
return key_caches, value_caches
706
707
708
+@lru_cache
709
+def print_info_once(msg: str) -> None:
710
+ # Set the stacklevel to 2 to print the caller's line info
711
+ logger.info(msg, stacklevel=2)
712
+
713
+
714
@lru_cache
715
def print_warning_once(msg: str) -> None:
716
# Set the stacklevel to 2 to print the caller's line info