Repository navigation
Expand file tree
/
Copy pathenvs.py
More file actions
executable file
·1255 lines (1069 loc) · 51.1 KB
/
Copy pathenvs.py
File metadata and controls
executable file
·1255 lines (1069 loc) · 51.1 KB
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
# SPDX-License-Identifier: Apache-2.0
# SPDX-FileCopyrightText: Copyright contributors to the vLLM project
import hashlib
import os
import sys
import tempfile
from typing import TYPE_CHECKING, Any, Callable, Optional
if TYPE_CHECKING:
VLLM_HOST_IP: str = ""
VLLM_PORT: Optional[int] = None
VLLM_RPC_BASE_PATH: str = tempfile.gettempdir()
VLLM_USE_MODELSCOPE: bool = False
VLLM_RINGBUFFER_WARNING_INTERVAL: int = 60
VLLM_NCCL_SO_PATH: Optional[str] = None
LD_LIBRARY_PATH: Optional[str] = None
VLLM_USE_TRITON_FLASH_ATTN: bool = True
VLLM_V1_USE_PREFILL_DECODE_ATTENTION: bool = False
VLLM_USE_AITER_UNIFIED_ATTENTION: bool = False
VLLM_FLASH_ATTN_VERSION: Optional[int] = None
LOCAL_RANK: int = 0
CUDA_VISIBLE_DEVICES: Optional[str] = None
VLLM_ENGINE_ITERATION_TIMEOUT_S: int = 60
VLLM_API_KEY: Optional[str] = None
S3_ACCESS_KEY_ID: Optional[str] = None
S3_SECRET_ACCESS_KEY: Optional[str] = None
S3_ENDPOINT_URL: Optional[str] = None
VLLM_MODEL_REDIRECT_PATH: Optional[str] = None
VLLM_CACHE_ROOT: str = os.path.expanduser("~/.cache/vllm")
VLLM_CONFIG_ROOT: str = os.path.expanduser("~/.config/vllm")
VLLM_USAGE_STATS_SERVER: str = "https://stats.vllm.ai"
VLLM_NO_USAGE_STATS: bool = False
VLLM_DO_NOT_TRACK: bool = False
VLLM_USAGE_SOURCE: str = ""
VLLM_CONFIGURE_LOGGING: int = 1
VLLM_LOGGING_LEVEL: str = "INFO"
VLLM_LOGGING_PREFIX: str = ""
VLLM_LOGGING_CONFIG_PATH: Optional[str] = None
VLLM_LOGITS_PROCESSOR_THREADS: Optional[int] = None
VLLM_LOG_STATS_INTERVAL: float = 10.
VLLM_TRACE_FUNCTION: int = 0
VLLM_ATTENTION_BACKEND: Optional[str] = None
VLLM_USE_FLASHINFER_SAMPLER: Optional[bool] = None
VLLM_PP_LAYER_PARTITION: Optional[str] = None
VLLM_CPU_KVCACHE_SPACE: Optional[int] = 0
VLLM_CPU_OMP_THREADS_BIND: str = ""
VLLM_CPU_NUM_OF_RESERVED_CPU: Optional[int] = None
VLLM_CPU_MOE_PREPACK: bool = True
VLLM_CPU_SGL_KERNEL: bool = False
VLLM_XLA_CACHE_PATH: str = os.path.join(VLLM_CACHE_ROOT, "xla_cache")
VLLM_XLA_CHECK_RECOMPILATION: bool = False
VLLM_FUSED_MOE_CHUNK_SIZE: int = 64 * 1024
VLLM_ENABLE_FUSED_MOE_ACTIVATION_CHUNKING: bool = True
VLLM_USE_RAY_SPMD_WORKER: bool = False
VLLM_USE_RAY_COMPILED_DAG: bool = False
VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE: str = "auto"
VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM: bool = False
VLLM_USE_RAY_WRAPPED_PP_COMM: bool = True
VLLM_XLA_USE_SPMD: bool = False
VLLM_WORKER_MULTIPROC_METHOD: str = "fork"
VLLM_ASSETS_CACHE: str = os.path.join(VLLM_CACHE_ROOT, "assets")
VLLM_IMAGE_FETCH_TIMEOUT: int = 5
VLLM_VIDEO_FETCH_TIMEOUT: int = 30
VLLM_AUDIO_FETCH_TIMEOUT: int = 10
VLLM_MEDIA_LOADING_THREAD_COUNT: int = 8
VLLM_MAX_AUDIO_CLIP_FILESIZE_MB: int = 25
VLLM_VIDEO_LOADER_BACKEND: str = "opencv"
VLLM_MM_INPUT_CACHE_GIB: int = 4
VLLM_TARGET_DEVICE: str = "cuda"
MAX_JOBS: Optional[str] = None
NVCC_THREADS: Optional[str] = None
VLLM_USE_PRECOMPILED: bool = False
VLLM_DOCKER_BUILD_CONTEXT: bool = False
VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL: bool = False
VLLM_KEEP_ALIVE_ON_ENGINE_DEATH: bool = False
CMAKE_BUILD_TYPE: Optional[str] = None
VERBOSE: bool = False
VLLM_ALLOW_LONG_MAX_MODEL_LEN: bool = False
VLLM_RPC_TIMEOUT: int = 10000 # ms
VLLM_HTTP_TIMEOUT_KEEP_ALIVE: int = 5 # seconds
VLLM_PLUGINS: Optional[list[str]] = None
VLLM_LORA_RESOLVER_CACHE_DIR: Optional[str] = None
VLLM_TORCH_PROFILER_DIR: Optional[str] = None
VLLM_TORCH_PROFILER_RECORD_SHAPES: bool = False
VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY: bool = False
VLLM_TORCH_PROFILER_WITH_STACK: bool = True
VLLM_TORCH_PROFILER_WITH_FLOPS: bool = False
VLLM_USE_TRITON_AWQ: bool = False
VLLM_ALLOW_RUNTIME_LORA_UPDATING: bool = False
VLLM_SKIP_P2P_CHECK: bool = False
VLLM_DISABLED_KERNELS: list[str] = []
VLLM_USE_V1: bool = True
VLLM_ROCM_USE_AITER: bool = False
VLLM_ROCM_USE_AITER_PAGED_ATTN: bool = False
VLLM_ROCM_USE_AITER_LINEAR: bool = True
VLLM_ROCM_USE_AITER_MOE: bool = True
VLLM_ROCM_USE_AITER_RMSNORM: bool = True
VLLM_ROCM_USE_AITER_MLA: bool = True
VLLM_ROCM_USE_AITER_MHA: bool = True
VLLM_ROCM_USE_SKINNY_GEMM: bool = True
VLLM_ROCM_FP8_PADDING: bool = True
VLLM_ROCM_MOE_PADDING: bool = True
VLLM_ROCM_CUSTOM_PAGED_ATTN: bool = True
VLLM_ENABLE_V1_MULTIPROCESSING: bool = True
VLLM_LOG_BATCHSIZE_INTERVAL: float = -1
VLLM_DISABLE_COMPILE_CACHE: bool = False
Q_SCALE_CONSTANT: int = 200
K_SCALE_CONSTANT: int = 200
V_SCALE_CONSTANT: int = 100
VLLM_SERVER_DEV_MODE: bool = False
VLLM_V1_OUTPUT_PROC_CHUNK_SIZE: int = 128
VLLM_MLA_DISABLE: bool = False
VLLM_RAY_PER_WORKER_GPUS: float = 1.0
VLLM_RAY_BUNDLE_INDICES: str = ""
VLLM_CUDART_SO_PATH: Optional[str] = None
VLLM_DP_RANK: int = 0
VLLM_DP_RANK_LOCAL: int = -1
VLLM_DP_SIZE: int = 1
VLLM_DP_MASTER_IP: str = ""
VLLM_DP_MASTER_PORT: int = 0
VLLM_MOE_DP_CHUNK_SIZE: int = 256
VLLM_RANDOMIZE_DP_DUMMY_INPUTS: bool = False
VLLM_MARLIN_USE_ATOMIC_ADD: bool = False
VLLM_MXFP4_USE_MARLIN: Optional[bool] = None
VLLM_V0_USE_OUTLINES_CACHE: bool = False
VLLM_V1_USE_OUTLINES_CACHE: bool = False
VLLM_TPU_BUCKET_PADDING_GAP: int = 0
VLLM_TPU_MOST_MODEL_LEN: Optional[int] = None
VLLM_TPU_USING_PATHWAYS: bool = False
VLLM_USE_DEEP_GEMM: bool = False
VLLM_USE_DEEP_GEMM_E8M0: bool = True
VLLM_SKIP_DEEP_GEMM_WARMUP: bool = False
VLLM_USE_FLASHINFER_MOE_FP8: bool = False
VLLM_USE_FLASHINFER_MOE_FP4: bool = False
VLLM_FLASHINFER_MOE_BACKEND: str = "throughput"
VLLM_XGRAMMAR_CACHE_MB: int = 0
VLLM_MSGPACK_ZERO_COPY_THRESHOLD: int = 256
VLLM_ALLOW_INSECURE_SERIALIZATION: bool = False
VLLM_NIXL_SIDE_CHANNEL_HOST: str = "localhost"
VLLM_NIXL_SIDE_CHANNEL_PORT: int = 5557
VLLM_ALL2ALL_BACKEND: str = "naive"
VLLM_MAX_TOKENS_PER_EXPERT_FP4_MOE: int = 163840
VLLM_TOOL_PARSE_REGEX_TIMEOUT_SECONDS: int = 1
VLLM_SLEEP_WHEN_IDLE: bool = False
VLLM_MQ_MAX_CHUNK_BYTES_MB: int = 16
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: int = 300
VLLM_KV_CACHE_LAYOUT: Optional[str] = None
VLLM_COMPUTE_NANS_IN_LOGITS: bool = False
VLLM_USE_NVFP4_CT_EMULATIONS: bool = False
VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: str = "NONE"
VLLM_ROCM_QUICK_REDUCE_CAST_BF16_TO_FP16: bool = True
VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB: Optional[int] = None
VLLM_NIXL_ABORT_REQUEST_TIMEOUT: int = 120
VLLM_USE_CUDNN_PREFILL: bool = False
VLLM_ENABLE_CUDAGRAPH_GC: bool = False
VLLM_LOOPBACK_IP: str = ""
VLLM_ALLOW_CHUNKED_LOCAL_ATTN_WITH_HYBRID_KV_CACHE: bool = False
VLLM_ENABLE_RESPONSES_API_STORE: bool = False
VLLM_USE_TRTLLM_ATTENTION: Optional[str] = None
VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8: bool = False
VLLM_USE_FLASHINFER_MOE_MXFP4_BF16: bool = False
VLLM_TUNED_CONFIG_FOLDER: Optional[str] = None
def get_default_cache_root():
return os.getenv(
"XDG_CACHE_HOME",
os.path.join(os.path.expanduser("~"), ".cache"),
)
def get_default_config_root():
return os.getenv(
"XDG_CONFIG_HOME",
os.path.join(os.path.expanduser("~"), ".config"),
)
def maybe_convert_int(value: Optional[str]) -> Optional[int]:
if value is None:
return None
return int(value)
def maybe_convert_bool(value: Optional[str]) -> Optional[bool]:
if value is None:
return None
return bool(int(value))
def get_vllm_port() -> Optional[int]:
"""Get the port from VLLM_PORT environment variable.
Returns:
The port number as an integer if VLLM_PORT is set, None otherwise.
Raises:
ValueError: If VLLM_PORT is a URI, suggest k8s service discovery issue.
"""
if 'VLLM_PORT' not in os.environ:
return None
port = os.getenv('VLLM_PORT', '0')
try:
return int(port)
except ValueError as err:
from urllib.parse import urlparse
parsed = urlparse(port)
if parsed.scheme:
raise ValueError(
f"VLLM_PORT '{port}' appears to be a URI. "
"This may be caused by a Kubernetes service discovery issue,"
"check the warning in: https://docs.vllm.ai/en/stable/serving/env_vars.html"
) from None
raise ValueError(
f"VLLM_PORT '{port}' must be a valid integer") from err
# The begin-* and end* here are used by the documentation generator
# to extract the used env vars.
# --8<-- [start:env-vars-definition]
environment_variables: dict[str, Callable[[], Any]] = {
# ================== Installation Time Env Vars ==================
# Target device of vLLM, supporting [cuda (by default),
# rocm, neuron, cpu]
"VLLM_TARGET_DEVICE":
lambda: os.getenv("VLLM_TARGET_DEVICE", "cuda").lower(),
# Maximum number of compilation jobs to run in parallel.
# By default this is the number of CPUs
"MAX_JOBS":
lambda: os.getenv("MAX_JOBS", None),
# Number of threads to use for nvcc
# By default this is 1.
# If set, `MAX_JOBS` will be reduced to avoid oversubscribing the CPU.
"NVCC_THREADS":
lambda: os.getenv("NVCC_THREADS", None),
# If set, vllm will use precompiled binaries (*.so)
"VLLM_USE_PRECOMPILED":
lambda: os.environ.get("VLLM_USE_PRECOMPILED", "").strip().lower() in
("1", "true") or bool(os.environ.get("VLLM_PRECOMPILED_WHEEL_LOCATION")),
# Used to mark that setup.py is running in a Docker build context,
# in order to force the use of precompiled binaries.
"VLLM_DOCKER_BUILD_CONTEXT":
lambda: os.environ.get("VLLM_DOCKER_BUILD_CONTEXT", "").strip().lower() in
("1", "true"),
# Whether to force using nightly wheel in python build.
# This is used for testing the nightly wheel in python build.
"VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL":
lambda: bool(int(os.getenv("VLLM_TEST_USE_PRECOMPILED_NIGHTLY_WHEEL", "0"))
),
# CMake build type
# If not set, defaults to "Debug" or "RelWithDebInfo"
# Available options: "Debug", "Release", "RelWithDebInfo"
"CMAKE_BUILD_TYPE":
lambda: os.getenv("CMAKE_BUILD_TYPE"),
# If set, vllm will print verbose logs during installation
"VERBOSE":
lambda: bool(int(os.getenv('VERBOSE', '0'))),
# Root directory for vLLM configuration files
# Defaults to `~/.config/vllm` unless `XDG_CONFIG_HOME` is set
# Note that this not only affects how vllm finds its configuration files
# during runtime, but also affects how vllm installs its configuration
# files during **installation**.
"VLLM_CONFIG_ROOT":
lambda: os.path.expanduser(
os.getenv(
"VLLM_CONFIG_ROOT",
os.path.join(get_default_config_root(), "vllm"),
)),
# ================== Runtime Env Vars ==================
# Root directory for vLLM cache files
# Defaults to `~/.cache/vllm` unless `XDG_CACHE_HOME` is set
"VLLM_CACHE_ROOT":
lambda: os.path.expanduser(
os.getenv(
"VLLM_CACHE_ROOT",
os.path.join(get_default_cache_root(), "vllm"),
)),
# used in distributed environment to determine the ip address
# of the current node, when the node has multiple network interfaces.
# If you are using multi-node inference, you should set this differently
# on each node.
'VLLM_HOST_IP':
lambda: os.getenv('VLLM_HOST_IP', ""),
# used in distributed environment to manually set the communication port
# Note: if VLLM_PORT is set, and some code asks for multiple ports, the
# VLLM_PORT will be used as the first port, and the rest will be generated
# by incrementing the VLLM_PORT value.
'VLLM_PORT':
get_vllm_port,
# path used for ipc when the frontend api server is running in
# multi-processing mode to communicate with the backend engine process.
'VLLM_RPC_BASE_PATH':
lambda: os.getenv('VLLM_RPC_BASE_PATH', tempfile.gettempdir()),
# If true, will load models from ModelScope instead of Hugging Face Hub.
# note that the value is true or false, not numbers
"VLLM_USE_MODELSCOPE":
lambda: os.environ.get("VLLM_USE_MODELSCOPE", "False").lower() == "true",
# Interval in seconds to log a warning message when the ring buffer is full
"VLLM_RINGBUFFER_WARNING_INTERVAL":
lambda: int(os.environ.get("VLLM_RINGBUFFER_WARNING_INTERVAL", "60")),
# path to cudatoolkit home directory, under which should be bin, include,
# and lib directories.
"CUDA_HOME":
lambda: os.environ.get("CUDA_HOME", None),
# Path to the NCCL library file. It is needed because nccl>=2.19 brought
# by PyTorch contains a bug: https://github.com/NVIDIA/nccl/issues/1234
"VLLM_NCCL_SO_PATH":
lambda: os.environ.get("VLLM_NCCL_SO_PATH", None),
# when `VLLM_NCCL_SO_PATH` is not set, vllm will try to find the nccl
# library file in the locations specified by `LD_LIBRARY_PATH`
"LD_LIBRARY_PATH":
lambda: os.environ.get("LD_LIBRARY_PATH", None),
# flag to control if vllm should use triton flash attention
"VLLM_USE_TRITON_FLASH_ATTN":
lambda: (os.environ.get("VLLM_USE_TRITON_FLASH_ATTN", "True").lower() in
("true", "1")),
# Use separate prefill and decode kernels for V1 attention instead of
# the unified triton kernel.
"VLLM_V1_USE_PREFILL_DECODE_ATTENTION":
lambda:
(os.getenv("VLLM_V1_USE_PREFILL_DECODE_ATTENTION", "False").lower() in
("true", "1")),
# Use AITER triton unified attention for V1 attention
"VLLM_USE_AITER_UNIFIED_ATTENTION":
lambda:
(os.getenv("VLLM_USE_AITER_UNIFIED_ATTENTION", "False").lower() in
("true", "1")),
# Force vllm to use a specific flash-attention version (2 or 3), only valid
# when using the flash-attention backend.
"VLLM_FLASH_ATTN_VERSION":
lambda: maybe_convert_int(os.environ.get("VLLM_FLASH_ATTN_VERSION", None)),
# Internal flag to enable Dynamo fullgraph capture
"VLLM_TEST_DYNAMO_FULLGRAPH_CAPTURE":
lambda: bool(
os.environ.get("VLLM_TEST_DYNAMO_FULLGRAPH_CAPTURE", "1") != "0"),
# Feature flag to enable/disable Inductor standalone compile.
# In torch <= 2.7 we ignore this flag; in torch >= 2.8 this is
# enabled by default.
"VLLM_USE_STANDALONE_COMPILE":
lambda: os.environ.get("VLLM_USE_STANDALONE_COMPILE", "1") == "1",
# local rank of the process in the distributed setting, used to determine
# the GPU device id
"LOCAL_RANK":
lambda: int(os.environ.get("LOCAL_RANK", "0")),
# used to control the visible devices in the distributed setting
"CUDA_VISIBLE_DEVICES":
lambda: os.environ.get("CUDA_VISIBLE_DEVICES", None),
# timeout for each iteration in the engine
"VLLM_ENGINE_ITERATION_TIMEOUT_S":
lambda: int(os.environ.get("VLLM_ENGINE_ITERATION_TIMEOUT_S", "60")),
# API key for vLLM API server
"VLLM_API_KEY":
lambda: os.environ.get("VLLM_API_KEY", None),
# Whether to log responses from API Server for debugging
"VLLM_DEBUG_LOG_API_SERVER_RESPONSE":
lambda: os.environ.get("VLLM_DEBUG_LOG_API_SERVER_RESPONSE", "False"
).lower() == "true",
# S3 access information, used for tensorizer to load model from S3
"S3_ACCESS_KEY_ID":
lambda: os.environ.get("S3_ACCESS_KEY_ID", None),
"S3_SECRET_ACCESS_KEY":
lambda: os.environ.get("S3_SECRET_ACCESS_KEY", None),
"S3_ENDPOINT_URL":
lambda: os.environ.get("S3_ENDPOINT_URL", None),
# Usage stats collection
"VLLM_USAGE_STATS_SERVER":
lambda: os.environ.get("VLLM_USAGE_STATS_SERVER", "https://stats.vllm.ai"),
"VLLM_NO_USAGE_STATS":
lambda: os.environ.get("VLLM_NO_USAGE_STATS", "0") == "1",
"VLLM_DO_NOT_TRACK":
lambda: (os.environ.get("VLLM_DO_NOT_TRACK", None) or os.environ.get(
"DO_NOT_TRACK", None) or "0") == "1",
"VLLM_USAGE_SOURCE":
lambda: os.environ.get("VLLM_USAGE_SOURCE", "production"),
# Logging configuration
# If set to 0, vllm will not configure logging
# If set to 1, vllm will configure logging using the default configuration
# or the configuration file specified by VLLM_LOGGING_CONFIG_PATH
"VLLM_CONFIGURE_LOGGING":
lambda: int(os.getenv("VLLM_CONFIGURE_LOGGING", "1")),
"VLLM_LOGGING_CONFIG_PATH":
lambda: os.getenv("VLLM_LOGGING_CONFIG_PATH"),
# this is used for configuring the default logging level
"VLLM_LOGGING_LEVEL":
lambda: os.getenv("VLLM_LOGGING_LEVEL", "INFO").upper(),
# if set, VLLM_LOGGING_PREFIX will be prepended to all log messages
"VLLM_LOGGING_PREFIX":
lambda: os.getenv("VLLM_LOGGING_PREFIX", ""),
# if set, vllm will call logits processors in a thread pool with this many
# threads. This is useful when using custom logits processors that either
# (a) launch additional CUDA kernels or (b) do significant CPU-bound work
# while not holding the python GIL, or both.
"VLLM_LOGITS_PROCESSOR_THREADS":
lambda: int(os.getenv("VLLM_LOGITS_PROCESSOR_THREADS", "0"))
if "VLLM_LOGITS_PROCESSOR_THREADS" in os.environ else None,
# If set, vllm will log stats at this interval in seconds
# If not set, vllm will log stats every 10 seconds.
"VLLM_LOG_STATS_INTERVAL":
lambda: val if (val := float(os.getenv("VLLM_LOG_STATS_INTERVAL", "10.")))
> 0. else 10.,
# Trace function calls
# If set to 1, vllm will trace function calls
# Useful for debugging
"VLLM_TRACE_FUNCTION":
lambda: int(os.getenv("VLLM_TRACE_FUNCTION", "0")),
# Backend for attention computation
# Available options:
# - "TORCH_SDPA": use torch.nn.MultiheadAttention
# - "FLASH_ATTN": use FlashAttention
# - "XFORMERS": use XFormers
# - "ROCM_FLASH": use ROCmFlashAttention
# - "FLASHINFER": use flashinfer
# - "FLASHMLA": use FlashMLA
"VLLM_ATTENTION_BACKEND":
lambda: os.getenv("VLLM_ATTENTION_BACKEND", None),
# If set, vllm will use flashinfer sampler
"VLLM_USE_FLASHINFER_SAMPLER":
lambda: bool(int(os.environ["VLLM_USE_FLASHINFER_SAMPLER"]))
if "VLLM_USE_FLASHINFER_SAMPLER" in os.environ else None,
# Pipeline stage partition strategy
"VLLM_PP_LAYER_PARTITION":
lambda: os.getenv("VLLM_PP_LAYER_PARTITION", None),
# (CPU backend only) CPU key-value cache space.
# default is None and will be set as 4 GB
"VLLM_CPU_KVCACHE_SPACE":
lambda: int(os.getenv("VLLM_CPU_KVCACHE_SPACE", "0"))
if "VLLM_CPU_KVCACHE_SPACE" in os.environ else None,
# (CPU backend only) CPU core ids bound by OpenMP threads, e.g., "0-31",
# "0,1,2", "0-31,33". CPU cores of different ranks are separated by '|'.
"VLLM_CPU_OMP_THREADS_BIND":
lambda: os.getenv("VLLM_CPU_OMP_THREADS_BIND", "auto"),
# (CPU backend only) CPU cores not used by OMP threads .
# Those CPU cores will not be used by OMP threads of a rank.
"VLLM_CPU_NUM_OF_RESERVED_CPU":
lambda: int(os.getenv("VLLM_CPU_NUM_OF_RESERVED_CPU", "0"))
if "VLLM_CPU_NUM_OF_RESERVED_CPU" in os.environ else None,
# (CPU backend only) whether to use prepack for MoE layer. This will be
# passed to ipex.llm.modules.GatedMLPMOE. On unsupported CPUs, you might
# need to set this to "0" (False).
"VLLM_CPU_MOE_PREPACK":
lambda: bool(int(os.getenv("VLLM_CPU_MOE_PREPACK", "1"))),
# (CPU backend only) whether to use SGL kernels, optimized for small batch.
"VLLM_CPU_SGL_KERNEL":
lambda: bool(int(os.getenv("VLLM_CPU_SGL_KERNEL", "0"))),
# If the env var is set, then all workers will execute as separate
# processes from the engine, and we use the same mechanism to trigger
# execution on all workers.
# Run vLLM with VLLM_USE_RAY_SPMD_WORKER=1 to enable it.
"VLLM_USE_RAY_SPMD_WORKER":
lambda: bool(int(os.getenv("VLLM_USE_RAY_SPMD_WORKER", "0"))),
# If the env var is set, it uses the Ray's Compiled Graph
# (previously known as ADAG) API which optimizes the
# control plane overhead.
# Run vLLM with VLLM_USE_RAY_COMPILED_DAG=1 to enable it.
# Note that this variable is set to 1 in V1 by default
# when ray distributed executor is used.
"VLLM_USE_RAY_COMPILED_DAG":
lambda: bool(int(os.getenv("VLLM_USE_RAY_COMPILED_DAG", "0"))),
# If the env var is set, Ray Compiled Graph uses the specified
# channel type to communicate between workers belonging to
# different pipeline-parallel stages.
# Available options:
# - "auto": use the default channel type
# - "nccl": use NCCL for communication
# - "shm": use shared memory and gRPC for communication
# This flag is ignored if VLLM_USE_RAY_COMPILED_DAG is not set.
"VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE":
lambda: os.getenv("VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE", "auto"),
# If the env var is set, it enables GPU communication overlap
# (experimental feature) in Ray's Compiled Graph. This flag is ignored if
# VLLM_USE_RAY_COMPILED_DAG is not set.
"VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM":
lambda: bool(int(os.getenv("VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM", "0"))
),
# If the env var is set, it uses a Ray Communicator wrapping
# vLLM's pipeline parallelism communicator to interact with Ray's
# Compiled Graph. Otherwise, it uses Ray's NCCL communicator.
# This flag is ignored if VLLM_USE_RAY_COMPILED_DAG is not set.
"VLLM_USE_RAY_WRAPPED_PP_COMM":
lambda: bool(int(os.getenv("VLLM_USE_RAY_WRAPPED_PP_COMM", "1"))),
# Use dedicated multiprocess context for workers.
# Both spawn and fork work
"VLLM_WORKER_MULTIPROC_METHOD":
lambda: os.getenv("VLLM_WORKER_MULTIPROC_METHOD", "fork"),
# Path to the cache for storing downloaded assets
"VLLM_ASSETS_CACHE":
lambda: os.path.expanduser(
os.getenv(
"VLLM_ASSETS_CACHE",
os.path.join(get_default_cache_root(), "vllm", "assets"),
)),
# Timeout for fetching images when serving multimodal models
# Default is 5 seconds
"VLLM_IMAGE_FETCH_TIMEOUT":
lambda: int(os.getenv("VLLM_IMAGE_FETCH_TIMEOUT", "5")),
# Timeout for fetching videos when serving multimodal models
# Default is 30 seconds
"VLLM_VIDEO_FETCH_TIMEOUT":
lambda: int(os.getenv("VLLM_VIDEO_FETCH_TIMEOUT", "30")),
# Timeout for fetching audio when serving multimodal models
# Default is 10 seconds
"VLLM_AUDIO_FETCH_TIMEOUT":
lambda: int(os.getenv("VLLM_AUDIO_FETCH_TIMEOUT", "10")),
# Max number of workers for the thread pool handling
# media bytes loading. Set to 1 to disable parallel processing.
# Default is 8
"VLLM_MEDIA_LOADING_THREAD_COUNT":
lambda: int(os.getenv("VLLM_MEDIA_LOADING_THREAD_COUNT", "8")),
# Maximum filesize in MB for a single audio file when processing
# speech-to-text requests. Files larger than this will be rejected.
# Default is 25 MB
"VLLM_MAX_AUDIO_CLIP_FILESIZE_MB":
lambda: int(os.getenv("VLLM_MAX_AUDIO_CLIP_FILESIZE_MB", "25")),
# Backend for Video IO
# - "opencv": Default backend that uses OpenCV stream buffered backend.
#
# Custom backend implementations can be registered
# via `@VIDEO_LOADER_REGISTRY.register("my_custom_video_loader")` and
# imported at runtime.
# If a non-existing backend is used, an AssertionError will be thrown.
"VLLM_VIDEO_LOADER_BACKEND":
lambda: os.getenv("VLLM_VIDEO_LOADER_BACKEND", "opencv"),
# [DEPRECATED] Cache size (in GiB per process) for multimodal input cache
# Default is 4 GiB per API process + 4 GiB per engine core process
"VLLM_MM_INPUT_CACHE_GIB":
lambda: int(os.getenv("VLLM_MM_INPUT_CACHE_GIB", "4")),
# Path to the XLA persistent cache directory.
# Only used for XLA devices such as TPUs.
"VLLM_XLA_CACHE_PATH":
lambda: os.path.expanduser(
os.getenv(
"VLLM_XLA_CACHE_PATH",
os.path.join(get_default_cache_root(), "vllm", "xla_cache"),
)),
# If set, assert on XLA recompilation after each execution step.
"VLLM_XLA_CHECK_RECOMPILATION":
lambda: bool(int(os.getenv("VLLM_XLA_CHECK_RECOMPILATION", "0"))),
# Enable SPMD mode for TPU backend.
"VLLM_XLA_USE_SPMD":
lambda: bool(int(os.getenv("VLLM_XLA_USE_SPMD", "0"))),
"VLLM_FUSED_MOE_CHUNK_SIZE":
lambda: int(os.getenv("VLLM_FUSED_MOE_CHUNK_SIZE", "32768")),
# Control whether to use fused MoE activation chunking. Current chunking
# logic is incompatible with torch.compile and causes IMA. See issue
# https://github.com/vllm-project/vllm/issues/19631.
"VLLM_ENABLE_FUSED_MOE_ACTIVATION_CHUNKING":
lambda: bool(
int(os.getenv("VLLM_ENABLE_FUSED_MOE_ACTIVATION_CHUNKING", "1"))),
# If set, the OpenAI API server will stay alive even after the underlying
# AsyncLLMEngine errors and stops serving requests
"VLLM_KEEP_ALIVE_ON_ENGINE_DEATH":
lambda: bool(os.getenv("VLLM_KEEP_ALIVE_ON_ENGINE_DEATH", 0)),
# If the env var VLLM_ALLOW_LONG_MAX_MODEL_LEN is set, it allows
# the user to specify a max sequence length greater than
# the max length derived from the model's config.json.
# To enable this, set VLLM_ALLOW_LONG_MAX_MODEL_LEN=1.
"VLLM_ALLOW_LONG_MAX_MODEL_LEN":
lambda:
(os.environ.get("VLLM_ALLOW_LONG_MAX_MODEL_LEN", "0").strip().lower() in
("1", "true")),
# If set, forces FP8 Marlin to be used for FP8 quantization regardless
# of the hardware support for FP8 compute.
"VLLM_TEST_FORCE_FP8_MARLIN":
lambda:
(os.environ.get("VLLM_TEST_FORCE_FP8_MARLIN", "0").strip().lower() in
("1", "true")),
"VLLM_TEST_FORCE_LOAD_FORMAT":
lambda: os.getenv("VLLM_TEST_FORCE_LOAD_FORMAT", "dummy"),
# Time in ms for the zmq client to wait for a response from the backend
# server for simple data operations
"VLLM_RPC_TIMEOUT":
lambda: int(os.getenv("VLLM_RPC_TIMEOUT", "10000")),
# Timeout in seconds for keeping HTTP connections alive in API server
"VLLM_HTTP_TIMEOUT_KEEP_ALIVE":
lambda: int(os.environ.get("VLLM_HTTP_TIMEOUT_KEEP_ALIVE"
10BC8
;, "5")),
# a list of plugin names to load, separated by commas.
# if this is not set, it means all plugins will be loaded
# if this is set to an empty string, no plugins will be loaded
"VLLM_PLUGINS":
lambda: None if "VLLM_PLUGINS" not in os.environ else os.environ[
"VLLM_PLUGINS"].split(","),
# a local directory to look in for unrecognized LoRA adapters.
# only works if plugins are enabled and
# VLLM_ALLOW_RUNTIME_LORA_UPDATING is enabled.
"VLLM_LORA_RESOLVER_CACHE_DIR":
lambda: os.getenv("VLLM_LORA_RESOLVER_CACHE_DIR", None),
# Enables torch profiler if set.
# Both AsyncLLM's CPU traces as well as workers'
# traces (CPU & GPU) will be saved under this directory.
# Note that it must be an absolute path.
"VLLM_TORCH_PROFILER_DIR":
lambda: (None if os.getenv("VLLM_TORCH_PROFILER_DIR", None) is None else os
.path.abspath(os.path.expanduser(os.getenv(
"VLLM_TORCH_PROFILER_DIR", ".")))),
# Enable torch profiler to record shapes if set
# VLLM_TORCH_PROFILER_RECORD_SHAPES=1. If not set, torch profiler will
# not record shapes.
"VLLM_TORCH_PROFILER_RECORD_SHAPES":
lambda: bool(os.getenv("VLLM_TORCH_PROFILER_RECORD_SHAPES", "0") != "0"),
# Enable torch profiler to profile memory if set
# VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY=1. If not set, torch profiler
# will not profile memory.
"VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY":
lambda: bool(
os.getenv("VLLM_TORCH_PROFILER_WITH_PROFILE_MEMORY", "0") != "0"),
# Enable torch profiler to profile stack if set
# VLLM_TORCH_PROFILER_WITH_STACK=1. If not set, torch profiler WILL
# profile stack by default.
"VLLM_TORCH_PROFILER_WITH_STACK":
lambda: bool(os.getenv("VLLM_TORCH_PROFILER_WITH_STACK", "1") != "0"),
# Enable torch profiler to profile flops if set
# VLLM_TORCH_PROFILER_WITH_FLOPS=1. If not set, torch profiler will
# not profile flops.
"VLLM_TORCH_PROFILER_WITH_FLOPS":
lambda: bool(os.getenv("VLLM_TORCH_PROFILER_WITH_FLOPS", "0") != "0"),
# If set, vLLM will use Triton implementations of AWQ.
"VLLM_USE_TRITON_AWQ":
lambda: bool(int(os.getenv("VLLM_USE_TRITON_AWQ", "0"))),
# If set, allow loading or unloading lora adapters in runtime,
"VLLM_ALLOW_RUNTIME_LORA_UPDATING":
lambda:
(os.environ.get("VLLM_ALLOW_RUNTIME_LORA_UPDATING", "0").strip().lower() in
("1", "true")),
# We assume drivers can report p2p status correctly.
# If the program hangs when using custom allreduce,
# potantially caused by a bug in the driver (535 series),
# if might be helpful to set VLLM_SKIP_P2P_CHECK=0
# so that vLLM can verify if p2p is actually working.
# See https://github.com/vllm-project/vllm/blob/a9b15c606fea67a072416ea0ea115261a2756058/vllm/distributed/device_communicators/custom_all_reduce_utils.py#L101-L108 for details. # noqa
"VLLM_SKIP_P2P_CHECK":
lambda: os.getenv("VLLM_SKIP_P2P_CHECK", "1") == "1",
# List of quantization kernels that should be disabled, used for testing
# and performance comparisons. Currently only affects MPLinearKernel
# selection
# (kernels: MacheteLinearKernel, MarlinLinearKernel, ExllamaLinearKernel)
"VLLM_DISABLED_KERNELS":
lambda: [] if "VLLM_DISABLED_KERNELS" not in os.environ else os.environ[
"VLLM_DISABLED_KERNELS"].split(","),
# If set, use the V1 code path.
"VLLM_USE_V1":
lambda: bool(int(os.getenv("VLLM_USE_V1", "1"))),
# Disable aiter ops unless specifically enabled.
# Acts as a parent switch to enable the rest of the other operations.
"VLLM_ROCM_USE_AITER":
lambda: (os.getenv("VLLM_ROCM_USE_AITER", "False").lower() in
("true", "1")),
# Whether to use aiter paged attention.
# By default is disabled.
"VLLM_ROCM_USE_AITER_PAGED_ATTN":
lambda: (os.getenv("VLLM_ROCM_USE_AITER_PAGED_ATTN", "False").lower() in
("true", "1")),
# use aiter linear op if aiter ops are enabled
# The following list of related ops
# - scaled_mm (per-tensor / rowwise)
"VLLM_ROCM_USE_AITER_LINEAR":
lambda: (os.getenv("VLLM_ROCM_USE_AITER_LINEAR", "True").lower() in
("true", "1")),
# Whether to use aiter moe ops.
# By default is enabled.
"VLLM_ROCM_USE_AITER_MOE":
lambda: (os.getenv("VLLM_ROCM_USE_AITER_MOE", "True").lower() in
("true", "1")),
# use aiter rms norm op if aiter ops are enabled.
"VLLM_ROCM_USE_AITER_RMSNORM":
lambda: (os.getenv("VLLM_ROCM_USE_AITER_RMSNORM", "True").lower() in
("true", "1")),
# Whether to use aiter mla ops.
# By default is enabled.
"VLLM_ROCM_USE_AITER_MLA":
lambda: (os.getenv("VLLM_ROCM_USE_AITER_MLA", "True").lower() in
("true", "1")),
# Whether to use aiter mha ops.
# By default is enabled.
"VLLM_ROCM_USE_AITER_MHA":
lambda: (os.getenv("VLLM_ROCM_USE_AITER_MHA", "True").lower() in
("true", "1")),
# use rocm skinny gemms
"VLLM_ROCM_USE_SKINNY_GEMM":
lambda: (os.getenv("VLLM_ROCM_USE_SKINNY_GEMM", "True").lower() in
("true", "1")),
# Pad the fp8 weights to 256 bytes for ROCm
"VLLM_ROCM_FP8_PADDING":
lambda: bool(int(os.getenv("VLLM_ROCM_FP8_PADDING", "1"))),
# Pad the weights for the moe kernel
"VLLM_ROCM_MOE_PADDING":
lambda: bool(int(os.getenv("VLLM_ROCM_MOE_PADDING", "1"))),
# custom paged attention kernel for MI3* cards
"VLLM_ROCM_CUSTOM_PAGED_ATTN":
lambda: (os.getenv("VLLM_ROCM_CUSTOM_PAGED_ATTN", "True").lower() in
("true", "1")),
# Custom quick allreduce kernel for MI3* cards
# Choice of quantization level: FP, INT8, INT6, INT4 or NONE
# Recommended for large models to get allreduce
"VLLM_ROCM_QUICK_REDUCE_QUANTIZATION":
lambda: os.getenv("VLLM_ROCM_QUICK_REDUCE_QUANTIZATION", "NONE").upper(),
# Custom quick allreduce kernel for MI3* cards
# Due to the lack of the bfloat16 asm instruction, bfloat16
# kernels are slower than fp16,
# If environment variable is set to 1, the input is converted to fp16
"VLLM_ROCM_QUICK_REDUCE_CAST_BF16_TO_FP16":
lambda:
(os.getenv("VLLM_ROCM_QUICK_REDUCE_CAST_BF16_TO_FP16", "True").lower() in
("true", "1")),
# Custom quick allreduce kernel for MI3* cards.
# Controls the maximum allowed number of data bytes(MB) for custom quick
# allreduce communication.
# Default: 2048 MB.
# Data exceeding this size will use either custom allreduce or RCCL
# communication.
"VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB":
lambda: maybe_convert_int(
os.environ.get("VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB", None)),
# Divisor for dynamic query scale factor calculation for FP8 KV Cache
"Q_SCALE_CONSTANT":
lambda: int(os.getenv("Q_SCALE_CONSTANT", "200")),
# Divisor for dynamic key scale factor calculation for FP8 KV Cache
"K_SCALE_CONSTANT":
lambda: int(os.getenv("K_SCALE_CONSTANT", "200")),
# Divisor for dynamic value scale factor calculation for FP8 KV Cache
"V_SCALE_CONSTANT":
lambda: int(os.getenv("V_SCALE_CONSTANT", "100")),
# If set, enable multiprocessing in LLM for the V1 code path.
"VLLM_ENABLE_V1_MULTIPROCESSING":
lambda: bool(int(os.getenv("VLLM_ENABLE_V1_MULTIPROCESSING", "1"))),
"VLLM_LOG_BATCHSIZE_INTERVAL":
lambda: float(os.getenv("VLLM_LOG_BATCHSIZE_INTERVAL", "-1")),
"VLLM_DISABLE_COMPILE_CACHE":
lambda: bool(int(os.getenv("VLLM_DISABLE_COMPILE_CACHE", "0"))),
# If set, vllm will run in development mode, which will enable
# some additional endpoints for developing and debugging,
# e.g. `/reset_prefix_cache`
"VLLM_SERVER_DEV_MODE":
lambda: bool(int(os.getenv("VLLM_SERVER_DEV_MODE", "0"))),
# Controls the maximum number of requests to handle in a
# single asyncio task when processing per-token outputs in the
# V1 AsyncLLM interface. It is applicable when handling a high
# concurrency of streaming requests.
# Setting this too high can result in a higher variance of
# inter-message latencies. Setting it too low can negatively impact
# TTFT and overall throughput.
"VLLM_V1_OUTPUT_PROC_CHUNK_SIZE":
lambda: int(os.getenv("VLLM_V1_OUTPUT_PROC_CHUNK_SIZE", "128")),
# If set, vLLM will disable the MLA attention optimizations.
"VLLM_MLA_DISABLE":
lambda: bool(int(os.getenv("VLLM_MLA_DISABLE", "0"))),
# Number of GPUs per worker in Ray, if it is set to be a fraction,
# it allows ray to schedule multiple actors on a single GPU,
# so that users can colocate other actors on the same GPUs as vLLM.
"VLLM_RAY_PER_WORKER_GPUS":
lambda: float(os.getenv("VLLM_RAY_PER_WORKER_GPUS", "1.0")),
# Bundle indices for Ray, if it is set, it can control precisely
# which indices are used for the Ray bundle, for every worker.
# Format: comma-separated list of integers, e.g. "0,1,2,3"
"VLLM_RAY_BUNDLE_INDICES":
lambda: os.getenv("VLLM_RAY_BUNDLE_INDICES", ""),
# In some system, find_loaded_library() may not work. So we allow users to
# specify the path through environment variable VLLM_CUDART_SO_PATH.
"VLLM_CUDART_SO_PATH":
lambda: os.getenv("VLLM_CUDART_SO_PATH", None),
# Rank of the process in the data parallel setting
"VLLM_DP_RANK":
lambda: int(os.getenv("VLLM_DP_RANK", "0")),
# Rank of the process in the data parallel setting.
# Defaults to VLLM_DP_RANK when not set.
"VLLM_DP_RANK_LOCAL":
lambda: int(
os.getenv("VLLM_DP_RANK_LOCAL", sys.modules[__name__].VLLM_DP_RANK)),
# World size of the data parallel setting
"VLLM_DP_SIZE":
lambda: int(os.getenv("VLLM_DP_SIZE", "1")),
# IP address of the master node in the data parallel setting
"VLLM_DP_MASTER_IP":
lambda: os.getenv("VLLM_DP_MASTER_IP", "127.0.0.1"),
# Port of the master node in the data parallel setting
"VLLM_DP_MASTER_PORT":
lambda: int(os.getenv("VLLM_DP_MASTER_PORT", "0")),
# In the context of executing MoE models with Data-Parallel, Expert-Parallel
# and Batched All-to-All dispatch/combine kernels, VLLM_MOE_DP_CHUNK_SIZE
# dictates the quantum of tokens that can be dispatched from a DP
# rank. All DP ranks process the activations in VLLM_MOE_DP_CHUNK_SIZE
# units.
"VLLM_MOE_DP_CHUNK_SIZE":
lambda: int(os.getenv("VLLM_MOE_DP_CHUNK_SIZE", "256")),
# Randomize inputs during dummy runs when using Data Parallel
"VLLM_RANDOMIZE_DP_DUMMY_INPUTS":
lambda: os.environ.get("VLLM_RANDOMIZE_DP_DUMMY_INPUTS", "0") == "1",
# Whether to use S3 path for model loading in CI via RunAI Streamer
"VLLM_CI_USE_S3":
lambda: os.environ.get("VLLM_CI_USE_S3", "0") == "1",
# Use model_redirect to redirect the model name to a local folder.
# `model_redirect` can be a json file mapping the model between
# repo_id and local folder:
# {"meta-llama/Llama-3.2-1B": "/tmp/Llama-3.2-1B"}
# or a space separated values table file:
# meta-llama/Llama-3.2-1B /tmp/Llama-3.2-1B
"VLLM_MODEL_REDIRECT_PATH":
lambda: os.environ.get("VLLM_MODEL_REDIRECT_PATH", None),
# Whether to use atomicAdd reduce in gptq/awq marlin kernel.
"VLLM_MARLIN_USE_ATOMIC_ADD":
lambda: os.environ.get("VLLM_MARLIN_USE_ATOMIC_ADD", "0") == "1",
# Whether to use marlin kernel in mxfp4 quantization method
"VLLM_MXFP4_USE_MARLIN":
lambda: maybe_convert_bool(os.environ.get("VLLM_MXFP4_USE_MARLIN", None)),
# Whether to turn on the outlines cache for V0
# This cache is unbounded and on disk, so it's not safe to use in
# an environment with potentially malicious users.
"VLLM_V0_USE_OUTLINES_CACHE":
lambda: os.environ.get("VLLM_V0_USE_OUTLINES_CACHE", "0") == "1",
# Whether to turn on the outlines cache for V1
# This cache is unbounded and on disk, so it's not safe to use in
# an environment with potentially malicious users.
"VLLM_V1_USE_OUTLINES_CACHE":
lambda: os.environ.get("VLLM_V1_USE_OUTLINES_CACHE", "0") == "1",
# Gap between padding buckets for the forward pass. So we have
# 8, we will run forward pass with [16, 24, 32, ...].
"VLLM_TPU_BUCKET_PADDING_GAP":
lambda: int(os.environ["VLLM_TPU_BUCKET_PADDING_GAP"])
if "VLLM_TPU_BUCKET_PADDING_GAP" in os.environ else 0,
"VLLM_TPU_MOST_MODEL_LEN":
lambda: maybe_convert_int(os.environ.get("VLLM_TPU_MOST_MODEL_LEN", None)),
# Whether using Pathways
"VLLM_TPU_USING_PATHWAYS":
lambda: bool("proxy" in os.getenv("JAX_PLATFORMS", "").lower()),
# Allow use of DeepGemm kernels for fused moe ops.
"VLLM_USE_DEEP_GEMM":
lambda: bool(int(os.getenv("VLLM_USE_DEEP_GEMM", "0"))),
# Whether to use E8M0 scaling when DeepGEMM is used on Blackwell GPUs.
# E8M0 is faster on B200 but may reduce accuracy.
"VLLM_USE_DEEP_GEMM_E8M0":
lambda: bool(int(os.getenv("VLLM_USE_DEEP_GEMM_E8M0", "1"))),
# DeepGemm JITs the kernels on-demand. The warmup attempts to make DeepGemm
# JIT all the required kernels before model execution so there is no
# JIT'ing in the hot-path. However, this warmup increases the engine
# startup time by a couple of minutes.
# Set `VLLM_SKIP_DEEP_GEMM_WARMUP` to disable the warmup.
"VLLM_SKIP_DEEP_GEMM_WARMUP":
lambda: bool(int(os.getenv("VLLM_SKIP_DEEP_GEMM_WARMUP", "0"))),
# Allow use of FlashInfer MoE kernels for fused moe ops.
"VLLM_USE_FLASHINFER_MOE_FP8":
lambda: bool(int(os.getenv("VLLM_USE_FLASHINFER_MOE_FP8", "0"))),
# Allow use of FlashInfer CUTLASS kernels for fused moe ops.
"VLLM_USE_FLASHINFER_MOE_FP4":
lambda: bool(int(os.getenv("VLLM_USE_FLASHINFER_MOE_FP4", "0"))),
# If set to 1, use the FlashInfer
# MXFP8 (activation) x MXFP4 (weight) MoE backend.
"VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8":
lambda: bool(int(os.getenv("VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8", "0"))),
# If set to 1, use the FlashInfer
# BF16 (activation) x MXFP4 (weight) MoE backend.
"VLLM_USE_FLASHINFER_MOE_MXFP4_BF16":
lambda: bool(int(os.getenv("VLLM_USE_FLASHINFER_MOE_MXFP4_BF16", "0"))),
# Control the cache sized used by the xgrammar compiler. The default
# of 512 MB should be enough for roughly 1000 JSON schemas.
# It can be changed with this variable if needed for some reason.
"VLLM_XGRAMMAR_CACHE_MB":
lambda: int(os.getenv("VLLM_XGRAMMAR_CACHE_MB", "512")),
# Control the threshold for msgspec to use 'zero copy' for
# serialization/deserialization of tensors. Tensors below
# this limit will be encoded into the msgpack buffer, and
# tensors above will instead be sent via a separate message.
# While the sending side still actually copies the tensor
# in all cases, on the receiving side, tensors above this
# limit will actually be zero-copy decoded.
"VLLM_MSGPACK_ZERO_COPY_THRESHOLD":
lambda: int(os.getenv("VLLM_MSGPACK_ZERO_COPY_THRESHOLD", "256")),
# If set, allow insecure serialization using pickle.
# This is useful for environments where it is deemed safe to use the
# insecure method and it is needed for some reason.