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 12, 18:30 UTC
[Deprecation] Deprecate `seed=None` (#29185)
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
Cyrus Leung committed
Nov 22, 2025 at 02:19 UTC
ceca06050124a10b33e78ee33d1a25a97edd1f74
1 file changed
+14
-5
vllm/engine/arg_utils.py
+14
-5
@@ -367,7 +367,7 @@ class EngineArgs:
367
config_format: str = ModelConfig.config_format
368
dtype: ModelDType = ModelConfig.dtype
369
kv_cache_dtype: CacheDType = CacheConfig.cache_dtype
370
- seed: int | None = None
370
+ seed: int | None = 0
371
max_model_len: int | None = ModelConfig.max_model_len
372
cuda_graph_sizes: list[int] | None = CompilationConfig.cudagraph_capture_sizes
373
cudagraph_capture_sizes: list[int] | None = (
@@ -1192,6 +1192,12 @@ class EngineArgs:
1192
# VLLM_ENABLE_V1_MULTIPROCESSING=0), so setting a seed here
1193
# doesn't affect the user process.
1194
if self.seed is None:
1195
+ logger.warning_once(
1196
+ "`seed=None` is equivalent to `seed=0` in V1 Engine. "
1197
+ "You will no longer be allowed to pass `None` in v0.13.",
1198
+ scope="local",
1199
+ )
1200
+
1201
self.seed = 0
1202
if not envs.VLLM_ENABLE_V1_MULTIPROCESSING:
1203
logger.warning(
@@ -1203,28 +1209,31 @@ class EngineArgs:
1209
)
1210
1211
if self.disable_mm_preprocessor_cache:
1206
- logger.warning(
1212
+ logger.warning_once(
1213
"`--disable-mm-preprocessor-cache` is deprecated "
1214
"and will be removed in v0.13. "
1215
"Please use `--mm-processor-cache-gb 0` instead.",
1216
+ scope="local",
1217
)
1218
1219
self.mm_processor_cache_gb = 0
1220
elif envs.VLLM_MM_INPUT_CACHE_GIB != 4:
1214
- logger.warning(
1221
+ logger.warning_once(
1222
"VLLM_MM_INPUT_CACHE_GIB` is deprecated "
1223
"and will be removed in v0.13. "
1224
"Please use `--mm-processor-cache-gb %d` instead.",
1225
envs.VLLM_MM_INPUT_CACHE_GIB,
1226
+ scope="local",
1227
)
1228
1229
self.mm_processor_cache_gb = envs.VLLM_MM_INPUT_CACHE_GIB
1230
1231
if self.enable_multimodal_encoder_data_parallel:
1224
- logger.warning(
1232
+ logger.warning_once(
1233
"--enable-multimodal-encoder-data-parallel` is deprecated "
1234
"and will be removed in v0.13. "
1227
- "Please use `--mm-encoder-tp-mode data` instead."
1235
+ "Please use `--mm-encoder-tp-mode data` instead.",
1236
+ scope="local",
1237
)
1238
1239
self.mm_encoder_tp_mode = "data"