[Log] Update to log once (#46511)
Signed-off-by: yewentao256 <zhyanwentao@126.com> Co-authored-by: TJian <tunjian.tan@embeddedllm.com>
Wentao Ye committed
Jun 24, 2026 at 10:45 UTC
007b5a52edf8e81a018be1df35164f98b730db9f
4 files changed
+23
-21
vllm/platforms/cpu.py
+7
-7
@@ -154,7 +154,7 @@ class CpuPlatform(Platform):
154
parallel_config.worker_cls = "vllm.v1.worker.cpu_worker.CPUWorker"
155
# Disable DBO
156
if parallel_config.enable_dbo:
157
- logger.warning("Dual-Batch Overlap is not supported on CPU, disabled.")
157
+ logger.warning_once("Dual-Batch Overlap is not supported on CPU, disabled.")
158
parallel_config.enable_dbo = False
159
160
if torch.cpu._is_amx_tile_supported() and (
@@ -166,7 +166,7 @@ class CpuPlatform(Platform):
166
):
167
cache_config.enable_prefix_caching = False
168
scheduler_config.enable_chunked_prefill = False
169
- logger.warning(
169
+ logger.warning_once(
170
"Disabled unsupported prefix caching and chunked prefill "
171
"for linear attention on AMX CPU platforms."
172
)
@@ -309,7 +309,7 @@ class CpuPlatform(Platform):
309
)
310
311
if model_config is not None and model_config.use_mla:
312
- logger.info(
312
+ logger.info_once(
313
"MLA is enabled on a non-GPU platform; forcing chunked "
314
"prefill and prefix caching to be disabled."
315
)
@@ -431,13 +431,13 @@ class CpuPlatform(Platform):
431
try:
432
import vllm._C # noqa: F401
433
except ImportError as e:
434
- logger.warning("Failed to import from vllm._C: %r", e)
434
+ logger.warning_once("Failed to import from vllm._C: %r", e)
435
else:
436
try:
437
import vllm._C_AVX512 # noqa: F401
438
except ImportError as e:
439
if ignored_msg not in e.msg:
440
- logger.warning(
440
+ logger.warning_once(
441
"Failed to import from vllm._C_AVX512: %r", e
442
)
443
else:
@@ -445,12 +445,12 @@ class CpuPlatform(Platform):
445
import vllm._C_AVX2 # noqa: F401
446
except ImportError as e:
447
if ignored_msg not in e.msg:
448
- logger.warning("Failed to import from vllm._C_AVX2: %r", e)
448
+ logger.warning_once("Failed to import from vllm._C_AVX2: %r", e)
449
else:
450
try:
451
import vllm._C # noqa: F401
452
except ImportError as e:
453
- logger.warning("Failed to import from vllm._C: %r", e)
453
+ logger.warning_once("Failed to import from vllm._C: %r", e)
454
455
@classmethod
456
def pack_kv_cache(
vllm/platforms/cuda.py
+8
-6
@@ -220,15 +220,17 @@ class CudaPlatformBase(Platform):
220
try:
221
import vllm._C_stable_libtorch # noqa: F401
222
except ImportError as e:
223
- logger.warning("Failed to import from vllm._C_stable_libtorch: %r", e)
223
+ logger.warning_once("Failed to import from vllm._C_stable_libtorch: %r", e)
224
try:
225
import vllm._moe_C_stable_libtorch # noqa: F401
226
except ImportError as e:
227
- logger.warning("Failed to import from vllm._moe_C_stable_libtorch: %r", e)
227
+ logger.warning_once(
228
+ "Failed to import from vllm._moe_C_stable_libtorch: %r", e
229
+ )
230
try:
231
import vllm._qutlass_C # noqa: F401
232
except ImportError as e:
231
- logger.warning("Failed to import from vllm._qutlass_C: %r", e)
233
+ logger.warning_once("Failed to import from vllm._qutlass_C: %r", e)
234
235
@property
236
def supported_dtypes(self) -> list[torch.dtype]:
@@ -291,7 +293,7 @@ class CudaPlatformBase(Platform):
293
# kernel with limited pinned memory support for CUDA.
294
version = _get_wsl_kernel_version()
295
if version is None or version < (4, 19, 121):
294
- logger.warning(
296
+ logger.warning_once(
297
"Using 'pin_memory=False' as WSL is detected and the "
298
"WSL2 kernel version is below 4.19.121. This may slow "
299
"down performance. Please run `wsl --update`."
@@ -320,7 +322,7 @@ class CudaPlatformBase(Platform):
322
and scheduler_config.is_multimodal_model
323
and not scheduler_config.disable_chunked_mm_input
324
):
323
- logger.warning(
325
+ logger.warning_once(
326
"Forcing --disable_chunked_mm_input for models "
327
"with multimodal-bidirectional attention."
328
)
@@ -331,7 +333,7 @@ class CudaPlatformBase(Platform):
333
and vllm_config.offload_config.uva.cpu_offload_gb > 0
334
and bool(vllm_config.compilation_config.cudagraph_mode)
335
):
334
- logger.warning(
336
+ logger.warning_once(
337
"--cpu-offload-gb is enabled with CUDA graphs on WSL2. "
338
"This combination requires pinned (page-locked) memory "
339
"allocations. WARNING: Windows (WDDM) enforces a hard "
vllm/platforms/interface.py
+3
-3
@@ -355,7 +355,7 @@ class Platform:
355
try:
356
import vllm._C # noqa: F401
357
except ImportError as e:
358
- logger.warning("Failed to import from vllm._C: %r", e)
358
+ logger.warning_once("Failed to import from vllm._C: %r", e)
359
with contextlib.suppress(ImportError):
360
import vllm._moe_C_stable_libtorch # noqa: F401
361
@@ -859,7 +859,7 @@ class Platform:
859
# Pinned memory support under WSL depends on the vendor and driver
860
# version. Conservative default: return False. Platform subclasses
861
# that can verify support (e.g. CudaPlatformBase) override this.
862
- logger.warning(
862
+ logger.warning_once(
863
"Using 'pin_memory=False' as WSL is detected. "
864
"This may slow down performance."
865
)
@@ -1003,7 +1003,7 @@ class Platform:
1003
if attr is not None:
1004
return attr
1005
1006
- logger.warning(
1006
+ logger.warning_once(
1007
"Current platform %s does not have '%s' attribute.",
1008
self.device_type,
1009
key,
vllm/platforms/xpu.py
+5
-5
@@ -55,7 +55,7 @@ class XPUPlatform(Platform):
55
from vllm.v1.attention.backends.utils import set_kv_cache_layout
56
57
set_kv_cache_layout("NHD")
58
- logger.info(
58
+ logger.info_once(
59
"Setting VLLM_KV_CACHE_LAYOUT to 'NHD' for XPU; "
60
"only NHD layout is supported by XPU attention kernels."
61
)
@@ -91,7 +91,7 @@ class XPUPlatform(Platform):
91
f"with use_mla: {attn_selector_config.use_mla}"
92
)
93
94
- logger.info("Using Flash Attention backend.")
94
+ logger.info_once("Using Flash Attention backend.")
95
return AttentionBackendEnum.FLASH_ATTN.get_path()
96
97
@classmethod
@@ -193,13 +193,13 @@ class XPUPlatform(Platform):
193
194
if not supports_xpu_graph():
195
compilation_config.cudagraph_mode = CUDAGraphMode.NONE
196
- logger.warning(
196
+ logger.warning_once(
197
"XPU Graph is not supported in the current PyTorch version, "
198
"disabling cudagraph_mode."
199
)
200
elif not envs.VLLM_XPU_ENABLE_XPU_GRAPH:
201
compilation_config.cudagraph_mode = CUDAGraphMode.NONE
202
- logger.warning(
202
+ logger.warning_once(
203
"XPU Graph is disabled by environment variable, "
204
"please set VLLM_XPU_ENABLE_XPU_GRAPH=1 to enable it."
205
)
@@ -218,7 +218,7 @@ class XPUPlatform(Platform):
218
if compilation_config.mode != CompilationMode.NONE:
219
for flag, feature_name in fusion_passes_to_disable.items():
220
if getattr(pass_config, flag):
221
- logger.warning(
221
+ logger.warning_once(
222
"Feature %r is not yet supported on XPU and will be disabled.",
223
feature_name,
224
)