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 6, 17:00 UTC
[Frontend] Reduce chat template warmup logging levels (#37062)
Signed-off-by: Nick Hill <nickhill123@gmail.com>
Nick Hill committed
Mar 14, 2026 at 13:48 UTC
458c1a4b2d21965ecd41b76ec0506ffe5ed8c8a1
1 file changed
+7
-8
vllm/renderers/base.py
+7
-8
@@ -179,17 +179,17 @@ class BaseRenderer(ABC, Generic[_T]):
179
from vllm.entrypoints.chat_utils import ChatTemplateResolutionError
180
181
try:
182
- logger.info("Warming up chat template processing...")
182
+ logger.debug("Warming up chat template processing...")
183
start_time = time.perf_counter()
184
185
self.render_chat([[{"role": "user", "content": "warmup"}]], chat_params)
186
187
elapsed = time.perf_counter() - start_time
188
- logger.info("Chat template warmup completed in %.3fs", elapsed)
188
+ logger.debug("Chat template warmup completed in %.3fs", elapsed)
189
except ChatTemplateResolutionError:
190
- logger.info("This model does not support chat template.")
190
+ logger.debug("This model does not support chat template.")
191
except Exception:
192
- logger.exception("Chat template warmup failed")
192
+ logger.warning("Chat template warmup failed", exc_info=True)
193
194
if self.mm_processor:
195
from vllm.multimodal.processing import TimingContext
@@ -200,7 +200,7 @@ class BaseRenderer(ABC, Generic[_T]):
200
mm_limits = processor.info.allowed_mm_limits
201
202
try:
203
- logger.info("Warming up multi-modal processing...")
203
+ logger.debug("Warming up multi-modal processing...")
204
start_time = time.perf_counter()
205
206
processor_inputs = processor.dummy_inputs.get_dummy_processor_inputs(
@@ -209,14 +209,13 @@ class BaseRenderer(ABC, Generic[_T]):
209
mm_options=mm_config.limit_per_prompt,
210
)
211
_ = processor.apply(
212
- processor_inputs,
213
- timing_ctx=TimingContext(enabled=False),
212
+ processor_inputs, timing_ctx=TimingContext(enabled=False)
213
)
214
215
elapsed = time.perf_counter() - start_time
216
logger.info("Multi-modal warmup completed in %.3fs", elapsed)
217
except Exception:
219
- logger.exception("Multi-modal warmup failed")
218
+ logger.warning("Multi-modal warmup failed")
219
finally:
220
self.clear_mm_cache()
221