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, 19:00 UTC
[log] enable max_log_len trim only when needed (#31482)
Signed-off-by: Andy Xie <andy.xning@gmail.com>
Ning Xie committed
Jan 5, 2026 at 11:55 UTC
98b8b3abaa26c8f593621b0230c37cd9d8b69015
1 file changed
+17
-15
vllm/entrypoints/logger.py
+17
-15
@@ -1,6 +1,7 @@
1
# SPDX-License-Identifier: Apache-2.0
2
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
3
4
+import logging
5
from collections.abc import Sequence
6
7
import torch
@@ -26,23 +27,24 @@ class RequestLogger:
27
params: SamplingParams | PoolingParams | BeamSearchParams | None,
28
lora_request: LoRARequest | None,
29
) -> None:
29
- max_log_len = self.max_log_len
30
- if max_log_len is not None:
31
- if prompt is not None:
32
- prompt = prompt[:max_log_len]
30
+ if logger.isEnabledFor(logging.DEBUG):
31
+ max_log_len = self.max_log_len
32
+ if max_log_len is not None:
33
+ if prompt is not None:
34
+ prompt = prompt[:max_log_len]
35
34
- if prompt_token_ids is not None:
35
- prompt_token_ids = prompt_token_ids[:max_log_len]
36
+ if prompt_token_ids is not None:
37
+ prompt_token_ids = prompt_token_ids[:max_log_len]
38
37
- logger.debug(
38
- "Request %s details: prompt: %r, "
39
- "prompt_token_ids: %s, "
40
- "prompt_embeds shape: %s.",
41
- request_id,
42
- prompt,
43
- prompt_token_ids,
44
- prompt_embeds.shape if prompt_embeds is not None else None,
45
- )
39
+ logger.debug(
40
+ "Request %s details: prompt: %r, "
41
+ "prompt_token_ids: %s, "
42
+ "prompt_embeds shape: %s.",
43
+ request_id,
44
+ prompt,
45
+ prompt_token_ids,
46
+ prompt_embeds.shape if prompt_embeds is not None else None,
47
+ )
48
49
logger.info(
50
"Received request %s: params: %s, lora_request: %s.",