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 15, 03:30 UTC
[Misc] Add logger.exception for TPU information collection failures (#20510)
Signed-off-by: reidliu41 <reid201711@gmail.com>
Reid committed
Jul 5, 2025 at 22:24 UTC
cf4cd53982161cb67c86b69f4a319625b231b0bd
1 file changed
+4
-1
vllm/usage/usage_lib.py
+4
-1
@@ -20,9 +20,12 @@ import torch
20
21
import vllm.envs as envs
22
from vllm.connections import global_http_connection
23
+from vllm.logger import init_logger
24
from vllm.utils import cuda_device_count_stateless, cuda_get_device_properties
25
from vllm.version import __version__ as VLLM_VERSION
26
27
+logger = init_logger(__name__)
28
+
29
_config_home = envs.VLLM_CONFIG_ROOT
30
_USAGE_STATS_JSON_PATH = os.path.join(_config_home, "usage_stats.json")
31
_USAGE_STATS_DO_NOT_TRACK_PATH = os.path.join(_config_home, "do_not_track")
@@ -183,7 +186,7 @@ class UsageMessage:
186
self.gpu_memory_per_device = (
187
torch_xla.core.xla_model.get_memory_info()["bytes_limit"])
188
except Exception:
186
- pass
189
+ logger.exception("Failed to collect TPU information")
190
self.provider = _detect_cloud_provider()
191
self.architecture = platform.machine()
192
self.platform = platform.platform()