Repository navigation
Expand file tree
/
Copy pathenvs.py
More file actions
executable file
·2410 lines (2305 loc) · 112 KB
/
Copy pathenvs.py
File metadata and controls
executable file
·2410 lines (2305 loc) · 112 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 functools
import json
import logging
import os
import sys
import tempfile
import uuid
from collections.abc import Callable
from typing import TYPE_CHECKING, Any, Literal
if TYPE_CHECKING:
VLLM_HOST_IP: str = ""
VLLM_PORT: int | None = None
VLLM_RPC_BASE_PATH: str = tempfile.gettempdir()
VLLM_USE_MODELSCOPE: bool = False
VLLM_USE_FASTOKENS: bool = False
VLLM_RINGBUFFER_WARNING_INTERVAL: int = 60
VLLM_NCCL_SO_PATH: str | None = None
LD_LIBRARY_PATH: str | None = None
VLLM_ROCM_SLEEP_MEM_CHUNK_SIZE: int = 256
LOCAL_RANK: int = 0
CUDA_VISIBLE_DEVICES: str | None = None
VLLM_ENGINE_ITERATION_TIMEOUT_S: int = 60
VLLM_ENGINE_READY_TIMEOUT_S: int = 600
VLLM_API_KEY: str | None = None
VLLM_DEBUG_LOG_API_SERVER_RESPONSE: bool = False
S3_ACCESS_KEY_ID: str | None = None
S3_SECRET_ACCESS_KEY: str | None = None
S3_ENDPOINT_URL: str | None = None
VLLM_MODEL_REDIRECT_PATH: str | None = None
VLLM_CACHE_ROOT: str = os.path.expanduser("~/.cache/vllm")
VLLM_CONFIG_ROOT: str = os.path.expanduser("~/.config/vllm")
VLLM_USAGE_STATS_SERVER: str =
8
"h
10BC8
ttps://stats.vllm.ai"
VLLM_NO_USAGE_STATS: bool = False
VLLM_DO_NOT_TRACK: bool = False
VLLM_USAGE_SOURCE: str = "production"
VLLM_CONFIGURE_LOGGING: bool = True
VLLM_LOGGING_LEVEL: str = "INFO"
VLLM_LOGGING_PREFIX: str = ""
VLLM_LOGGING_STREAM: str = "ext://sys.stdout"
VLLM_LOGGING_CONFIG_PATH: str | None = None
VLLM_LOGGING_COLOR: str = "auto"
NO_COLOR: bool = False
FORCE_COLOR: bool = False
VLLM_LOG_STATS_INTERVAL: float = 10.0
VLLM_TRACE_FUNCTION: int = 0
VLLM_USE_FLASHINFER_SAMPLER: bool = True
VLLM_PP_LAYER_PARTITION: str | None = None
VLLM_CPU_KVCACHE_SPACE: int | None = 0
VLLM_CPU_OMP_THREADS_BIND: str = "auto"
VLLM_CPU_NUM_OF_RESERVED_CPU: int | None = None
VLLM_CPU_CI_ENV: bool = False
VLLM_CPU_ATTN_SPLIT_KV: bool = True
VLLM_ZENTORCH_WEIGHT_PREPACK: bool = True
VLLM_CPU_INT4_W4A8: bool = True
VLLM_XLA_CACHE_PATH: str = os.path.join(VLLM_CACHE_ROOT, "xla_cache")
VLLM_XLA_CHECK_RECOMPILATION: bool = False
VLLM_SPARSE_INDEXER_MAX_LOGITS_MB: int = 512
VLLM_ADAPTIVE_VERIFICATION_PROFILE_CONTEXT_LEN: int = 8192
VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE: Literal["auto", "nccl", "shm"] = "auto"
VLLM_USE_RAY_COMPILED_DAG_OVERLAP_COMM: bool = False
VLLM_USE_RAY_WRAPPED_PP_COMM: bool = True
VLLM_USE_RAY_V2_EXECUTOR_BACKEND: bool = False
VLLM_DISTRIBUTED_USE_SPLIT_GROUP: bool = False
VLLM_XLA_USE_SPMD: bool = False
VLLM_WORKER_MULTIPROC_METHOD: Literal["fork", "spawn"] = "fork"
VLLM_ASSETS_CACHE: str = os.path.join(VLLM_CACHE_ROOT, "assets")
VLLM_ASSETS_CACHE_MODEL_CLEAN: bool = False
VLLM_IMAGE_FETCH_TIMEOUT: int = 5
VLLM_VIDEO_FETCH_TIMEOUT: int = 30
VLLM_AUDIO_FETCH_TIMEOUT: int = 10
VLLM_MEDIA_CACHE: str = ""
VLLM_MEDIA_CACHE_MAX_SIZE_MB: int = 5120
VLLM_MEDIA_CACHE_TTL_HOURS: float = 24
VLLM_MEDIA_FETCH_MAX_RETRIES: int = 3
VLLM_MAX_MEDIA_DOWNLOAD_SIZE_MB: int = 256
VLLM_MEDIA_URL_ALLOW_REDIRECTS: bool = True
VLLM_MEDIA_LOADING_THREAD_COUNT: int = 8
VLLM_MAX_AUDIO_CLIP_FILESIZE_MB: int = 25
VLLM_MAX_AUDIO_DECODE_DURATION_S: int = 600
VLLM_MAX_AUDIO_DECODE_BYTES: int = 268_435_456
VLLM_MAX_AUDIO_PREPROCESS_WORKERS: int = max(1, min(os.cpu_count() or 1, 2))
VLLM_MAX_EMBED_DECODE_BYTES: int = 2_147_483_648
VLLM_MAX_IMAGE_PIXELS: int = 178_956_970
VLLM_VIDEO_LOADER_BACKEND: str = "opencv"
VLLM_MEDIA_CONNECTOR: str = "http"
VLLM_TARGET_DEVICE: str = "cuda"
VLLM_MAIN_CUDA_VERSION: str = "13.0"
VLLM_FLOAT32_MATMUL_PRECISION: Literal["highest", "high", "medium"] = "highest"
VLLM_BATCH_INVARIANT: bool = False
VLLM_TRITON_USE_TD: bool | None = None
VLLM_GPU_SYNC_CHECK: Literal["warn", "error"] | None = None
MAX_JOBS: str | None = None
NVCC_THREADS: str | None = None
VLLM_USE_PRECOMPILED: bool = False
VLLM_USE_PRECOMPILED_RUST: bool = False
VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX: bool = False
VLLM_DOCKER_BUILD_CONTEXT: bool = False
VLLM_BUILD_COMMIT: str = "unknown"
VLLM_BUILD_PIPELINE: str = "local"
VLLM_BUILD_URL: str = ""
VLLM_IMAGE_TAG: str = ""
VLLM_KEEP_ALIVE_ON_ENGINE_DEATH: bool = False
CMAKE_BUILD_TYPE: Literal["Debug", "Release", "RelWithDebInfo"] | None = None
VERBOSE: bool = False
VLLM_ALLOW_LONG_MAX_MODEL_LEN: bool = False
VLLM_HTTP_TIMEOUT_KEEP_ALIVE: int = 5 # seconds
VLLM_MAX_N_SEQUENCES: int = 16384
VLLM_MAX_COMPLETION_PROMPTS: int = 1024
VLLM_MAX_STOP_STRINGS: int = 4
VLLM_MAX_NUM_BAD_WORDS: int = 128
VLLM_MAX_BAD_WORDS_TOTAL_TOKENS: int = 1024
VLLM_PLUGINS: list[str] | None = None
VLLM_LORA_RESOLVER_CACHE_DIR: str | None = None
VLLM_LORA_RESOLVER_HF_REPO_LIST: str | None = None
VLLM_USE_AOT_COMPILE: bool = False
VLLM_USE_BYTECODE_HOOK: bool = True
VLLM_FORCE_AOT_LOAD: bool = False
VLLM_USE_MEGA_AOT_ARTIFACT: bool = False
VLLM_USE_TRITON_AWQ: bool = False
VLLM_FASTSAFETENSORS_QUEUE_SIZE: int = 0
VLLM_TRITON_FORCE_FIRST_CONFIG: bool = False
VLLM_ALLOW_RUNTIME_LORA_UPDATING: bool = False
VLLM_SKIP_P2P_CHECK: bool = False
VLLM_DISABLED_KERNELS: list[str] = []
VLLM_USE_HW_AGNOSTIC: bool = False
VLLM_ENABLE_FLA_PACKED_RECURRENT_DECODE: bool = True
VLLM_GDN_DECODE_KERNEL: Literal["cuda", "triton"] = "cuda"
VLLM_DISABLE_PYNCCL: bool = False
VLLM_USE_OINK_OPS: bool = False
VLLM_MXFP4_EMULATION_DEQUANT_AT_LOAD: bool = False
VLLM_MXFP8_EMULATION_DEQUANT_AT_LOAD: bool = True
VLLM_ROCM_USE_AITER: bool = False
VLLM_ROCM_USE_AITER_CUSTOM_AR: bool = True
VLLM_ROCM_USE_AITER_LINEAR: bool = True
VLLM_ROCM_USE_AITER_LINEAR_HIPBMM: bool = False
VLLM_ROCM_USE_AITER_MOE: bool = True
VLLM_ROCM_AITER_MOE_DISPATCH_POLICY: int = 0
VLLM_ROCM_USE_AITER_MOE_SITUV2: bool = False
VLLM_ROCM_USE_AITER_RMSNORM: bool = True
VLLM_ROCM_USE_AITER_MLA: bool = True
VLLM_ROCM_AITER_MLA_ASM_PADDING: Literal["auto", "gluon", "asm"] = "auto"
VLLM_ROCM_USE_AITER_MHA: bool = True
VLLM_ROCM_USE_AITER_FP4_ASM_GEMM: bool = False
VLLM_ROCM_USE_AITER_TRITON_ROPE: bool = False
VLLM_ROCM_USE_AITER_FP8BMM: bool = True
VLLM_ROCM_USE_AITER_FP4BMM: bool = True
VLLM_ROCM_USE_AITER_UNIFIED_ATTENTION: bool = False
VLLM_ROCM_USE_AITER_FUSION_SHARED_EXPERTS: bool = False
VLLM_ROCM_USE_AITER_TRITON_GEMM: bool = True
VLLM_ROCM_USE_SKINNY_GEMM: bool = True
VLLM_ROCM_FP8_PADDING: bool = True
VLLM_ROCM_MOE_PADDING: bool = True
VLLM_ROCM_SHUFFLE_KV_CACHE_LAYOUT: bool = False
VLLM_ENABLE_V1_MULTIPROCESSING: bool = True
VLLM_LOG_BATCHSIZE_INTERVAL: float = -1
VLLM_DISABLE_COMPILE_CACHE: bool = False
VLLM_REPLICATE_EMBED: bool = False
VLLM_USE_LAYERNAME: bool = True
Q_SCALE_CONSTANT: int = 200
K_SCALE_CONSTANT: int = 200
V_SCALE_CONSTANT: int = 100
VLLM_USE_RUST_FRONTEND: bool = False
VLLM_USE_RUST_BENCH: bool = False
VLLM_RUST_FRONTEND_PATH: str | None = "auto"
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: str | None = None
VLLM_DP_RANK: int = 0
VLLM_DP_RANK_LOCAL: int = -1
VLLM_DP_SIZE: int = 1
VLLM_USE_STANDALONE_COMPILE: bool = True
VLLM_ENABLE_PREGRAD_PASSES: bool = True
VLLM_USE_BREAKABLE_CUDAGRAPH: bool = False
VLLM_DP_MASTER_IP: str = ""
VLLM_DP_MASTER_PORT: int = 0
VLLM_RANDOMIZE_DP_DUMMY_INPUTS: bool = False
VLLM_RAY_DP_PACK_STRATEGY: Literal["strict", "fill", "span"] = "strict"
VLLM_RAY_DP_PLACEMENT_NODE_IPS: str = ""
VLLM_RAY_EXTRA_ENV_VAR_PREFIXES_TO_COPY: str = ""
VLLM_RAY_EXTRA_ENV_VARS_TO_COPY: str = ""
VLLM_MARLIN_USE_ATOMIC_ADD: bool = False
VLLM_MARLIN_INPUT_DTYPE: Literal["int8", "fp8"] | None = None
VLLM_HUMMING_ONLINE_QUANT_CONFIG: dict[str, Any] | None = None
VLLM_HUMMING_INPUT_QUANT_CONFIG: dict[str, Any] | None = None
VLLM_HUMMING_USE_F16_ACCUM: bool = False
VLLM_HUMMING_MOE_GEMM_TYPE: Literal["indexed", "grouped", "auto"] | None = None
VLLM_B12X_MOE_FP4_FORCE_A16: bool = False
VLLM_DEEPEPLL_NVFP4_DISPATCH: bool = False
VLLM_V1_USE_OUTLINES_CACHE: bool = False
VLLM_TPU_USING_PATHWAYS: bool = False
VLLM_USE_DEEP_GEMM: bool = True
VLLM_MOE_USE_DEEP_GEMM: bool = True
VLLM_USE_DEEP_GEMM_E8M0: bool = True
VLLM_USE_DEEP_GEMM_TMA_ALIGNED_SCALES: bool = True
VLLM_DCP_Q_REPLICATE: bool = False
VLLM_USE_DIRECT_DCP_A2A: bool | None = None
VLLM_USE_DIRECT_DCP_Q_GATHER: bool | None = None
VLLM_USE_DIRECT_DCP_KV_GATHER: bool | None = None
VLLM_DEEP_GEMM_WARMUP: Literal[
"skip",
"full",
"relax",
] = "relax"
VLLM_USE_FUSED_MOE_GROUPED_TOPK: bool = True
VLLM_MOE_SKIP_PADDING: bool = True
VLLM_KIMI_K3_SHARD_SP_SHARED_EXPERT: bool = False
VLLM_KIMI_K3_AUX_ATTN_RES_STREAM: bool = False
VLLM_KIMI_K3_GEMM_AR: bool = True
VLLM_ENABLE_GEMM_RS: bool = False
VLLM_BLOCKSCALE_FP8_GEMM_FLASHINFER: bool = True
VLLM_USE_FLASHINFER_MOE_INT4: bool = False
VLLM_FLASHINFER_AUTOTUNE_CACHE_DIR: str | None = None
VLLM_FLASHINFER_AUTOTUNE_SKIP_OPS: list[str] | None = None
VLLM_FLASHINFER_ALLREDUCE_BACKEND: Literal["auto", "trtllm", "mnnvl"] = "auto"
VLLM_FLASHINFER_WORKSPACE_BUFFER_SIZE: int = 394 * 1024 * 1024
VLLM_XGRAMMAR_CACHE_MB: int = 0
VLLM_REGEX_COMPILATION_TIMEOUT_S: int = 5
VLLM_MSGPACK_ZERO_COPY_THRESHOLD: int = 256
VLLM_ALLOW_INSECURE_SERIALIZATION: bool = False
VLLM_DISABLE_REQUEST_ID_RANDOMIZATION: bool = False
VLLM_NIXL_SIDE_CHANNEL_HOST: str = "localhost"
VLLM_NIXL_SIDE_CHANNEL_PORT: int = 5600
VLLM_P2P_SIDE_CHANNEL_HOST: str = "localhost"
VLLM_P2P_SIDE_CHANNEL_PORT: int = 5710
VLLM_EC_SIDE_CHANNEL_HOST: str = "localhost"
VLLM_EC_SIDE_CHANNEL_PORT: int = 5601
VLLM_MOONCAKE_BOOTSTRAP_PORT: int = 8998
VLLM_MOONCAKE_STORE_TIER_LOG: bool = False
VLLM_MOONCAKE_LOAD_RECV_THREADS: int = 1
VLLM_MOONCAKE_DISK_STAGING_USABLE_RATIO: float = 0.9
MOONCAKE_PREFERRED_SEGMENT: str | None = None
MOONCAKE_REQUESTER_LOCAL_HOSTNAME: str | None = None
VLLM_MAX_TOKENS_PER_EXPERT_FP4_MOE: int = 163840
VLLM_TOOL_PARSE_REGEX_TIMEOUT_SECONDS: int = 1
VLLM_ENFORCE_STRICT_TOOL_CALLING: bool = True
VLLM_MQ_MAX_CHUNK_BYTES_MB: int = 16
VLLM_EXECUTE_MODEL_TIMEOUT_SECONDS: int = 300
VLLM_WORKER_SHUTDOWN_TIMEOUT_SECONDS: int = 5
VLLM_KV_CACHE_LAYOUT: (
Literal["LBNHC", "LBHNC", "LHBNC", "NHD", "HND", "BLHNC", "BLNHC", "BHLNC"]
| None
) = None
VLLM_SSM_CONV_STATE_LAYOUT: Literal["SD", "DS"] | None = None
VLLM_COMPUTE_NANS_IN_LOGITS: bool = False
VLLM_RAISE_ON_LOGIT_NANS: bool = False
VLLM_ROCM_QUICK_REDUCE_QUANTIZATION: Literal[
"FP", "INT8", "INT6", "INT4", "INT3", "NONE"
] = "NONE"
VLLM_ROCM_QUICK_REDUCE_CAST_BF16_TO_FP16: bool = True
VLLM_ROCM_QUICK_REDUCE_MAX_SIZE_BYTES_MB: int | None = None
VLLM_ROCM_QUICK_REDUCE_MIN_SIZE_BYTES_MB: int | None = None
VLLM_ROCM_QUICK_REDUCE_QUANTIZATION_MIN_SIZE_KB: int | None = None
VLLM_MOONCAKE_ABORT_REQUEST_TIMEOUT: int = 480
VLLM_ENABLE_CUDAGRAPH_GC: bool = False
VLLM_LOOPBACK_IP: str = ""
VLLM_ALLOW_CHUNKED_LOCAL_ATTN_WITH_HYBRID_KV_CACHE: bool = True
VLLM_ENABLE_RESPONSES_API_STORE: bool = False
VLLM_ENABLE_COHERE_API: bool = False
VLLM_HAS_FLASHINFER_CUBIN: bool = False
VLLM_ROCM_FP8_MFMA_PAGE_ATTN: bool = False
VLLM_ALLREDUCE_USE_SYMM_MEM: bool = True
VLLM_ALLREDUCE_USE_FLASHINFER: bool = True
VLLM_ALLREDUCE_USE_FLASHINFER_PCIE_IPC: bool = False
VLLM_TUNED_CONFIG_FOLDER: str | None = None
VLLM_ENABLE_STARTUP_PLAN: bool = False
VLLM_GPT_OSS_SYSTEM_TOOL_MCP_LABELS: set[str] = set()
VLLM_USE_EXPERIMENTAL_PARSER_CONTEXT: bool = False
VLLM_GPT_OSS_HARMONY_SYSTEM_INSTRUCTIONS: bool = False
VLLM_SYSTEM_START_DATE: str | None = None
VLLM_TOOL_JSON_ERROR_AUTOMATIC_RETRY: bool = False
VLLM_CUSTOM_SCOPES_FOR_PROFILING: bool = False
VLLM_NVTX_SCOPES_FOR_PROFILING: bool = False
VLLM_KV_EVENTS_USE_INT_BLOCK_HASHES: bool = True
VLLM_OBJECT_STORAGE_SHM_BUFFER_NAME: str = "VLLM_OBJECT_STORAGE_SHM_BUFFER"
VLLM_DEEPEP_BUFFER_SIZE_MB: int = 1024
VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE: bool = False
VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL: bool = False
VLLM_DEEPEP_V2_ALLOW_HYBRID_MODE: bool = True
VLLM_DEEPEP_V2_PREFER_OVERLAP: bool = False
VLLM_DEEPEP_V2_ALLOW_MULTIPLE_REDUCTION: bool = False
VLLM_DBO_COMM_SMS: int = 20
VLLM_PATTERN_MATCH_DEBUG: str | None = None
VLLM_DEBUG_DUMP_PATH: str | None = None
VLLM_ENABLE_INDUCTOR_MAX_AUTOTUNE: bool = True
VLLM_ENABLE_INDUCTOR_COORDINATE_DESCENT_TUNING: bool = True
VLLM_USE_NCCL_SYMM_MEM: bool = False
VLLM_NCCL_INCLUDE_PATH: str | None = None
VLLM_GC_DEBUG: str = ""
VLLM_DEBUG_WORKSPACE: bool = False
VLLM_DISABLE_SHARED_EXPERTS_STREAM: bool = False
VLLM_DISABLE_DSV4_MEGAMOE_SHARED_EXPERT_FUSION: bool = False
VLLM_SHARED_EXPERTS_STREAM_TOKEN_THRESHOLD: int = 256
VLLM_MULTI_STREAM_GEMM_TOKEN_THRESHOLD: int = 1024
VLLM_COMPILE_CACHE_SAVE_FORMAT: Literal["binary", "unpacked"] = "binary"
VLLM_USE_V2_MODEL_RUNNER: bool | None = None
VLLM_LOG_MODEL_INSPECTION: bool = False
VLLM_DEBUG_MFU_METRICS: bool = False
VLLM_WEIGHT_OFFLOADING_DISABLE_PIN_MEMORY: bool = False
VLLM_WEIGHT_OFFLOADING_DISABLE_UVA: bool = False
VLLM_KV_OFFLOAD_MAX_BATCH_DESCRIPTORS: int = 0
VLLM_WSL2_ENABLE_PIN_MEMORY: bool = False
VLLM_DISABLE_LOG_LOGO: bool = False
VLLM_LORA_DISABLE_PDL: bool = False
VLLM_ENABLE_CUDA_COMPATIBILITY: bool = False
VLLM_CUDA_COMPATIBILITY_PATH: str | None = None
VLLM_SKIP_MODEL_NAME_VALIDATION: bool = False
"""If set, vLLM will skip model name validation in API requests.
This allows any model name to be accepted in the 'model' field of requests,
making the server model-name agnostic. Useful for proxy/gateway scenarios."""
VLLM_ELASTIC_EP_SCALE_UP_LAUNCH: bool = False
VLLM_ELASTIC_EP_DRAIN_REQUESTS: bool = False
VLLM_MEMORY_PROFILER_ESTIMATE_CUDAGRAPHS: bool = True
VLLM_XPU_ENABLE_XPU_GRAPH: bool = False
VLLM_XPU_FORCE_N_CONTIG_WEIGHT: bool = False
VLLM_XPU_USE_SAMPLER_KERNEL: bool = True
VLLM_XPU_INC_WNA16_BACKEND: Literal["auto", "ark", "w4a16", "w4a8"] = "auto"
VLLM_LORA_ENABLE_DUAL_STREAM: bool = False
VLLM_GPU_NIC_PCIE_MAPPING: str = ""
VLLM_NIC_SELECTION_VARS: str = ""
VLLM_ENABLE_HPC_OPS: bool = False
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: str | None) -> int | None:
if value is None:
return None
return int(value)
def maybe_convert_bool(value: str | None) -> bool | None:
if value is None:
return None
return bool(int(value))
def maybe_convert_json_str_or_file(value: str | None) -> dict[str, Any] | None:
if value is None:
return None
if os.path.exists(value):
with open(value) as f:
return json.load(f)
return json.loads(value)
def disable_compile_cache() -> bool:
return bool(int(os.getenv("VLLM_DISABLE_COMPILE_CACHE", "0")))
def use_aot_compile() -> bool:
from vllm.utils.torch_utils import is_torch_equal_or_newer
default_value = (
"1"
if is_torch_equal_or_newer("2.10.0") and not disable_compile_cache()
else "0"
)
return os.environ.get("VLLM_USE_AOT_COMPILE", default_value) == "1"
def use_mega_aot_artifact():
from vllm.utils.torch_utils import is_torch_equal_or_newer
default_value = (
"1" if is_torch_equal_or_newer("2.12.0.dev") and use_aot_compile() else "0"
)
return os.environ.get("VLLM_USE_MEGA_AOT_ARTIFACT", default_value) == "1"
def env_with_choices(
env_name: str,
default: str | None,
choices: list[str] | Callable[[], list[str]],
case_sensitive: bool = True,
) -> Callable[[], str | None]:
"""Create a lambda that validates environment variable against allowed choices.
Args:
env_name: Name of the environment variable
default: Default value if not set (can be None)
choices: List of valid string options or callable that returns list
case_sensitive: Whether validation should be case sensitive
Returns:
Lambda function for environment_variables dict
"""
def _get_validated_env() -> str | None:
value = os.getenv(env_name)
if value is None:
return default
# Resolve choices if it's a callable (for lazy loading)
actual_choices = choices() if callable(choices) else choices
if not case_sensitive:
check_value = value.lower()
check_choices = [choice.lower() for choice in actual_choices]
else:
check_value = value
check_choices = actual_choices
if check_value not in check_choices:
raise ValueError(
f"Invalid value '{value}' for {env_name}. "
f"Valid options: {actual_choices}."
)
return value
return _get_validated_env
def env_list_with_choices(
env_name: str,
default: list[str],
choices: list[str] | Callable[[], list[str]],
case_sensitive: bool = True,
) -> Callable[[], list[str]]:
"""Create a lambda that validates environment variable
containing comma-separated values against allowed choices
Args:
env_name: Name of the environment variable
default: Default list of values if not set
choices: List of valid string options or callable that returns list
case_sensitive: Whether validation should be case sensitive
Returns:
Lambda function for environment_variables
dict that returns list of strings
"""
def _get_validated_env_list() -> list[str]:
value = os.getenv(env_name)
if value is None:
return default
# Split comma-separated values and strip whitespace
values = [v.strip() for v in value.split(",") if v.strip()]
if not values:
return default
# Resolve choices if it's a callable (for lazy loading)
actual_choices = choices() if callable(choices) else choices
# Validate each value
for val in values:
if not case_sensitive:
check_value = val.lower()
check_choices = [choice.lower() for choice in actual_choices]
else:
check_value = val
check_choices = actual_choices
if check_value not in check_choices:
raise ValueError(
f"Invalid value '{val}' in {env_name}. "
f"Valid options: {actual_choices}."
)
return values
return _get_validated_env_list
def env_set_with_choices(
env_name: str,
default: list[str],
choices: list[str] | Callable[[], list[str]],
case_sensitive: bool = True,
) -> Callable[[], set[str]]:
"""Creates a lambda which that validates environment variable
containing comma-separated values against allowed choices which
returns choices as a set.
"""
def _get_validated_env_set() -> set[str]:
return set(env_list_with_choices(env_name, default, choices, case_sensitive)())
return _get_validated_env_set
def get_vllm_port() -> int | None:
"""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 urllib3.util import parse_url
parsed = parse_url(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/latest/configuration/env_vars.html"
) from None
raise ValueError(f"VLLM_PORT '{port}' must be a valid integer") from err
def _generate_shm_name() -> str:
import pybase64
# Fit macOS's 30-character limit without truncating the UUID.
encoded_uuid = pybase64.urlsafe_b64encode(uuid.uuid4().bytes).decode("ascii")
return "vllm_mm_" + encoded_uuid.rstrip("=")
def get_env_or_set_default(
env_name: str,
default_factory: Callable[[], str],
) -> Callable[[], str]:
"""Create a lambda that returns an environment variable value if set,
or generates and sets a default value using the provided factory function.
"""
def _get_or_set_default() -> str:
value = os.getenv(env
10BC8
_name)
if value is not None:
return value
default_value = default_factory()
os.environ[env_name] = default_value
return default_value
return _get_or_set_default
# The start-* and end* here are used by the documentation generator
# to extract the used env vars.
# --8<-- [start:env-vars-definition]
logger = logging.getLogger(__name__)
def _resolve_rust_cli_path() -> str | None:
"""Resolve the vllm-rs binary path.
Returns None unless VLLM_USE_RUST_FRONTEND or VLLM_USE_RUST_BENCH is enabled.
When enabled, resolves VLLM_RUST_FRONTEND_PATH ("auto" by default)
to the actual binary path.
"""
use_rust = bool(int(os.environ.get("VLLM_USE_RUST_FRONTEND", "0"))) or bool(
int(os.environ.get("VLLM_USE_RUST_BENCH", "0"))
)
raw = os.environ.get("VLLM_RUST_FRONTEND_PATH", "auto")
if not use_rust:
if os.environ.get("VLLM_RUST_FRONTEND_PATH") is not None:
logger.warning(
"VLLM_RUST_FRONTEND_PATH is set without enabling "
"VLLM_USE_RUST_FRONTEND or VLLM_USE_RUST_BENCH. "
"Set one of them to 1 to use the vllm-rs binary."
)
return None
if raw.lower() in ("auto", "1", "true"):
pkg_dir = os.path.dirname(os.path.abspath(__file__))
candidate = os.path.join(pkg_dir, "vllm-rs")
if os.path.isfile(candidate) and os.access(candidate, os.X_OK):
return candidate
raise FileNotFoundError(
"VLLM_RUST_FRONTEND_PATH=auto but the vllm-rs binary was "
f"not found at {candidate}. "
"Build with setuptools-rust or set the path explicitly."
)
return raw
environment_variables: dict[str, Callable[[], Any]] = {
# ================== Installation Time Env Vars ==================
# Target device of vLLM, supporting [cuda (by default),
# rocm, cpu]
"VLLM_TARGET_DEVICE": lambda: os.getenv("VLLM_TARGET_DEVICE", "cuda").lower(),
# Main CUDA version of vLLM. This follows PyTorch but can be overridden.
"VLLM_MAIN_CUDA_VERSION": lambda: (
os.getenv("VLLM_MAIN_CUDA_VERSION", "").lower() or "13.0"
),
# Controls PyTorch float32 matmul precision mode within vLLM workers.
# Valid options mirror torch.set_float32_matmul_precision
"VLLM_FLOAT32_MATMUL_PRECISION": env_with_choices(
"VLLM_FLOAT32_MATMUL_PRECISION",
"highest",
["highest", "high", "medium"],
case_sensitive=False,
),
# Enable batch-invariant mode: deterministic results regardless of
# batch composition. Requires NVIDIA GPU with compute capability >= 9.0.
"VLLM_BATCH_INVARIANT": lambda: bool(int(os.getenv("VLLM_BATCH_INVARIANT", "0"))),
"VLLM_REPLICATE_EMBED": lambda: (
os.getenv("VLLM_REPLICATE_EMBED", "0").strip().lower() in ("1", "true")
),
# Use tensor descriptors for Q/K/V loads and output stores in the
# Triton unified-attention kernel. Enables HW 2D block reads on
# Intel XPU; the non-TD branch is dead-code-eliminated at Triton
# compile time so other platforms see no overhead. Tri-state override:
# unset (default) lets the `triton_attn` backend auto-select per
# platform (currently auto-enabled on XPU only); ``1`` forces TD on;
# ``0`` forces TD off. Useful for A/B benchmarking the TD path.
"VLLM_TRITON_USE_TD": lambda: {"1": True, "0": False}.get(
os.getenv("VLLM_TRITON_USE_TD", "").strip()
),
# If set, enable GPU<->CPU synchronization checking around the worker's
# `execute_model` and `sample_tokens` calls, via PyTorch's sync debug mode
# plus wrappers flagging `non_blocking` CPU<->CUDA copies that silently
# block the host (CPU tensors that are not pinned or not densely laid out).
# Valid values are "warn" (warn on each sync) or "error" (raise on sync).
# Unset disables the check. See `torch.cuda.set_sync_debug_mode`.
"VLLM_GPU_SYNC_CHECK": env_with_choices(
"VLLM_GPU_SYNC_CHECK", None, ["warn", "error"]
),
# 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 native binaries (*.so and vllm-rs).
"VLLM_USE_PRECOMPILED": lambda: (
os.environ.get("VLLM_USE_PRECOMPILED", "").strip().lower() in ("1", "true")
or bool(os.environ.get("VLLM_PRECOMPILED_WHEEL_LOCATION"))
),
# If set, vllm will use the precompiled Rust frontend binary (vllm-rs).
"VLLM_USE_PRECOMPILED_RUST": lambda: (
os.environ.get("VLLM_USE_PRECOMPILED_RUST", "").strip().lower() in ("1", "true")
),
# If set, skip adding +precompiled suffix to version string
"VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX": lambda: bool(
int(os.environ.get("VLLM_SKIP_PRECOMPILED_VERSION_SUFFIX", "0"))
),
# 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")
),
# Build provenance metadata embedded in official vllm-openai images.
# Set via Docker ENV at image build time; informational only.
"VLLM_BUILD_COMMIT": lambda: os.environ.get("VLLM_BUILD_COMMIT", "unknown"),
"VLLM_BUILD_PIPELINE": lambda: os.environ.get("VLLM_BUILD_PIPELINE", "local"),
"VLLM_BUILD_URL": lambda: os.environ.get("VLLM_BUILD_URL", ""),
"VLLM_IMAGE_TAG": lambda: os.environ.get("VLLM_IMAGE_TAG", ""),
# CMake build type
# If not set, defaults to "Debug" or "RelWithDebInfo"
# Available options: "Debug", "Release", "RelWithDebInfo"
"CMAKE_BUILD_TYPE": env_with_choices(
"CMAKE_BUILD_TYPE", None, ["Debug", "Release", "RelWithDebInfo"]
),
# 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.
"VLLM_USE_MODELSCOPE": lambda: (
os.environ.get("VLLM_USE_MODELSCOPE", "False").strip().lower() in ("1", "true")
),
# If true, replace the Rust BPE backend that powers HF fast tokenizers
# with the `fastokens` (https://github.com/crusoecloud/fastokens) shim.
# Applies to any tokenizer mode that loads an HF fast tokenizer
# (`hf`, `deepseek_v32`, `deepseek_v4`, …). The `fastokens`
# Python package must be installed.
"VLLM_USE_FASTOKENS": lambda: bool(int(os.getenv("VLLM_USE_FASTOKENS", "0"))),
# 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 the chunk size (in MB) for sleeping memory allocations under ROCm
"VLLM_ROCM_SLEEP_MEM_CHUNK_SIZE": lambda: int(
os.environ.get("VLLM_ROCM_SLEEP_MEM_CHUNK_SIZE", "256")
),
# Feature flag to enable/disable Inductor standalone compile.
# In torch <= 2.7 we ignore this flag; in torch >= 2.9 this is
# enabled by default.
"VLLM_USE_STANDALONE_COMPILE": lambda: (
os.environ.get("VLLM_USE_STANDALONE_COMPILE", "1") == "1"
),
# Inductor's pre-grad passes don't do anything for vLLM.
# The pre-grad passes get run even on cache-hit and negatively impact
# vllm cold compile times by O(1s)
# Can remove this after the following issue gets fixed
# TODO(luka): maybe_inplace requires this
# https://github.com/pytorch/pytorch/issues/174502
"VLLM_ENABLE_PREGRAD_PASSES": lambda: (
os.environ.get("VLLM_ENABLE_PREGRAD_PASSES", "1") == "1"
),
# Experimental: breakable cudagraph does not rely on torch.compile
"VLLM_USE_BREAKABLE_CUDAGRAPH": lambda: (
os.environ.get("VLLM_USE_BREAKABLE_CUDAGRAPH", "0") == "1"
),
# Debug pattern matching inside custom passes.
# Should be set to the fx.Node name (e.g. 'getitem_34' or 'scaled_mm_3').
"VLLM_PATTERN_MATCH_DEBUG": lambda: os.environ.get(
"VLLM_PATTERN_MATCH_DEBUG", None
),
# Dump fx graphs to the given directory.
# It will override CompilationConfig.debug_dump_path if set.
"VLLM_DEBUG_DUMP_PATH": lambda: os.environ.get("VLLM_DEBUG_DUMP_PATH", None),
# Feature flag to enable/disable AOT compilation. This will ensure
# compilation is done in warmup phase and the compilation will be
# reused in subsequent calls.
"VLLM_USE_AOT_COMPILE": use_aot_compile,
# Feature flag to enable/disable bytecode in
# TorchCompileWithNoGuardsWrapper.
"VLLM_USE_BYTECODE_HOOK": lambda: bool(
int(os.environ.get("VLLM_USE_BYTECODE_HOOK", "1"))
),
# Force vllm to always load AOT compiled models from disk. Failure
# to load will result in a hard error when this is enabled.
# Will be ignored when VLLM_USE_AOT_COMPILE is disabled.
"VLLM_FORCE_AOT_LOAD": lambda: os.environ.get("VLLM_FORCE_AOT_LOAD", "0") == "1",
# Enable loading compiled models directly from cached standalone compile artifacts
# without re-splitting graph modules. This reduces overhead during model
# loading by using reconstruct_serializable_fn_from_mega_artifact.
"VLLM_USE_MEGA_AOT_ARTIFACT": use_mega_aot_artifact,
# 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
108D6
in the engine
"VLLM_ENGINE_ITERATION_TIMEOUT_S": lambda: int(
os.environ.get("VLLM_ENGINE_ITERATION_TIMEOUT_S", "60")
),
# Timeout in seconds for waiting for engine cores to become ready
# during startup. Default is 600 seconds (10 minutes).
"VLLM_ENGINE_READY_TIMEOUT_S": lambda: int(
os.environ.get("VLLM_ENGINE_READY_TIMEOUT_S", "600")
),
# 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: bool(
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(),
# this is used for configuring the default logging stream
"VLLM_LOGGING_STREAM": lambda: os.getenv("VLLM_LOGGING_STREAM", "ext://sys.stdout"),
# if set, VLLM_LOGGING_PREFIX will be prepended to all log messages
"VLLM_LOGGING_PREFIX": lambda: os.getenv("VLLM_LOGGING_PREFIX", ""),
# Controls colored logging output. Options: "auto" (default, colors when terminal),
# "1" (always use colors), "0" (never use colors)
"VLLM_LOGGING_COLOR": lambda: os.getenv("VLLM_LOGGING_COLOR", "auto"),
# Standard unix flag for disabling ANSI color codes
"NO_COLOR": lambda: os.getenv("NO_COLOR", "0") != "0",
# De-facto standard flag for forcing ANSI color codes (e.g. non-tty case)
"FORCE_COLOR": lambda: os.getenv("FORCE_COLOR", "0") != "0",
# 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."))) > 0.0
else 10.0
),
# 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")),
# Whether to use the FlashInfer top-k / top-p sampler on CUDA. Enabled
# by default when the hardware supports it — set to 0 to opt out
# explicitly, which forces the PyTorch-native (Triton for bs>=8) path.
"VLLM_USE_FLASHINFER_SAMPLER": lambda: (
bool(int(os.environ["VLLM_USE_FLASHINFER_SAMPLER"]))
if "VLLM_USE_FLASHINFER_SAMPLER" in os.environ
else True
),
# 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 vLLM is running in a CI environment.
"VLLM_CPU_CI_ENV": lambda: bool(int(os.getenv("VLLM_CPU_CI_ENV", "0"))),
# (CPU backend only) whether to enable attention spilt KV.
"VLLM_CPU_ATTN_SPLIT_KV": lambda: bool(
int(os.getenv("VLLM_CPU_ATTN_SPLIT_KV", "1"))
),
# (Zen CPU backend) eagerly prepack weights into ZenDNN blocked layout
# at model load time. Eliminates per-inference layout conversion overhead.
"VLLM_ZENTORCH_WEIGHT_PREPACK": lambda: bool(
int(os.getenv("VLLM_ZENTORCH_WEIGHT_PREPACK", "1"))
),
# (CPU backend only) whether to use SGLang INT4 W4A8 kernels for AWQ.
"VLLM_CPU_INT4_W4A8": lambda: bool(int(os.getenv("VLLM_CPU_INT4_W4A8", "1"))),
# 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
"VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE": env_with_choices(
"VLLM_USE_RAY_COMPILED_DAG_CHANNEL_TYPE", "auto", ["auto", "nccl", "shm"]
),
# If the env var is set, it enables GPU communication overlap
# (experimental feature) in Ray's Compiled Graph.
"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.
"VLLM_USE_RAY_WRAPPED_PP_COMM": lambda: bool(
int(os.getenv("VLLM_USE_RAY_WRAPPED_PP_COMM", "1"))
),
# When True and distributed_executor_backend="ray", use RayExecutorV2
# (MQ-based) instead of RayDistributedExecutor (compiled-graph backend).
"VLLM_USE_RAY_V2_EXECUTOR_BACKEND": lambda: bool(
int(os.getenv("VLLM_USE_RAY_V2_EXECUTOR_BACKEND", "1"))
),
# When True, GroupCoordinator constructs its CPU/device subgroups via
# ``torch.distributed.split_group(backend=...)``
# and ``init_distributed_environment`` initializes the default PG with
# mixed ``cpu:gloo,cuda:nccl`` backend + eager ``device_id`` binding.
"VLLM_DISTRIBUTED_USE_SPLIT_GROUP": lambda: bool(
int(os.getenv("VLLM_DISTRIBUTED_USE_SPLIT_GROUP", "0"))
),
# Use dedicated multiprocess context for workers.
# Both spawn and fork work
"VLLM_WORKER_MULTIPROC_METHOD": env_with_choices(
"VLLM_WORKER_MULTIPROC_METHOD", "fork", ["spawn", "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"),
)
),
# If the env var is set, we will clean model file in
# this path $VLLM_ASSETS_CACHE/model_streamer/$model_name
"VLLM_ASSETS_CACHE_MODEL_CLEAN": lambda: bool(
int(os.getenv("VLLM_ASSETS_CACHE_MODEL_CLEAN", "0"))
),
# 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")
),
# Directory for caching media downloads (images, video, audio fetched
# from URLs during inference). Empty string disables caching.
"VLLM_MEDIA_CACHE": lambda: os.getenv("VLLM_MEDIA_CACHE", ""),
# Maximum cache size in MB. When exceeded, least-recently-used entries
# are evicted. Default is 5120 (5 GB).
"VLLM_MEDIA_CACHE_MAX_SIZE_MB": lambda: int(
os.getenv("VLLM_MEDIA_CACHE_MAX_SIZE_MB", "5120")
),
# Time-to-live in hours for cached media files. Entries older than this
# are evicted regardless of cache size. Default is 24 hours.
"VLLM_MEDIA_CACHE_TTL_HOURS": lambda: float(
os.getenv("VLLM_MEDIA_CACHE_TTL_HOURS", "24")
),
# Maximum number of retries for fetching media (images, audio, video)
# from URLs. Each retry quadruples the timeout. Default is 3.
"VLLM_MEDIA_FETCH_MAX_RETRIES": lambda: int(
os.getenv("VLLM_MEDIA_FETCH_MAX_RETRIES", "3")
),
# Maximum size in MB for a single remote media download. The limit is
# enforced while streaming the response body so oversized or infinite
# responses cannot grow the API server heap without bound. Default is 256.
"VLLM_MAX_MEDIA_DOWNLOAD_SIZE_MB": lambda: int(
os.getenv("VLLM_MAX_MEDIA_DOWNLOAD_SIZE_MB", "256")
),
# Whether to allow HTTP redirects when fetching from media URLs.
# Default to True
"VLLM_MEDIA_URL_ALLOW_REDIRECTS": lambda: bool(
int(os.getenv("VLLM_MEDIA_URL_ALLOW_REDIRECTS", "1"))
),
# Max number of workers for the thread pool handling
# media bytes loading. Set to 1 to disable parallel processing.
# Default is 8