sglang 0.5.2rc2__py3-none-any.whl → 0.5.3__py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (377) hide show
  1. sglang/bench_one_batch.py +7 -9
  2. sglang/bench_one_batch_server.py +330 -31
  3. sglang/bench_serving.py +267 -32
  4. sglang/global_config.py +2 -2
  5. sglang/lang/backend/runtime_endpoint.py +1 -1
  6. sglang/launch_server.py +14 -0
  7. sglang/profiler.py +2 -2
  8. sglang/srt/batch_invariant_ops/__init__.py +27 -0
  9. sglang/srt/batch_invariant_ops/batch_invariant_ops.py +549 -0
  10. sglang/srt/configs/__init__.py +8 -0
  11. sglang/srt/configs/device_config.py +3 -1
  12. sglang/srt/configs/dots_ocr.py +64 -0
  13. sglang/srt/configs/dots_vlm.py +139 -0
  14. sglang/srt/configs/falcon_h1.py +360 -0
  15. sglang/srt/configs/load_config.py +9 -0
  16. sglang/srt/configs/model_config.py +181 -82
  17. sglang/srt/configs/qwen3_next.py +326 -0
  18. sglang/srt/configs/qwen3_vl.py +586 -0
  19. sglang/srt/connector/__init__.py +8 -1
  20. sglang/srt/connector/remote_instance.py +82 -0
  21. sglang/srt/constrained/base_grammar_backend.py +49 -12
  22. sglang/srt/constrained/llguidance_backend.py +0 -1
  23. sglang/srt/constrained/outlines_backend.py +0 -1
  24. sglang/srt/constrained/outlines_jump_forward.py +1 -1
  25. sglang/srt/constrained/xgrammar_backend.py +30 -9
  26. sglang/srt/custom_op.py +11 -1
  27. sglang/srt/debug_utils/dump_comparator.py +81 -44
  28. sglang/srt/debug_utils/dump_loader.py +97 -0
  29. sglang/srt/debug_utils/dumper.py +21 -6
  30. sglang/srt/debug_utils/text_comparator.py +73 -11
  31. sglang/srt/disaggregation/ascend/conn.py +2 -2
  32. sglang/srt/disaggregation/ascend/transfer_engine.py +47 -9
  33. sglang/srt/disaggregation/base/conn.py +1 -1
  34. sglang/srt/disaggregation/common/conn.py +279 -108
  35. sglang/srt/disaggregation/decode.py +71 -19
  36. sglang/srt/disaggregation/decode_kvcache_offload_manager.py +185 -0
  37. sglang/srt/disaggregation/decode_schedule_batch_mixin.py +29 -17
  38. sglang/srt/disaggregation/fake/conn.py +1 -1
  39. sglang/srt/disaggregation/mini_lb.py +6 -445
  40. sglang/srt/disaggregation/mooncake/conn.py +55 -537
  41. sglang/srt/disaggregation/nixl/conn.py +326 -53
  42. sglang/srt/disaggregation/prefill.py +36 -17
  43. sglang/srt/disaggregation/utils.py +40 -54
  44. sglang/srt/distributed/device_communicators/all_reduce_utils.py +16 -0
  45. sglang/srt/distributed/device_communicators/shm_broadcast.py +4 -2
  46. sglang/srt/distributed/device_communicators/symm_mem.py +164 -0
  47. sglang/srt/distributed/parallel_state.py +156 -80
  48. sglang/srt/entrypoints/engine.py +59 -18
  49. sglang/srt/entrypoints/grpc_request_manager.py +855 -0
  50. sglang/srt/entrypoints/grpc_server.py +810 -0
  51. sglang/srt/entrypoints/http_server.py +130 -59
  52. sglang/srt/entrypoints/openai/protocol.py +112 -4
  53. sglang/srt/entrypoints/openai/serving_base.py +65 -3
  54. sglang/srt/entrypoints/openai/serving_chat.py +204 -55
  55. sglang/srt/entrypoints/openai/serving_completions.py +14 -3
  56. sglang/srt/entrypoints/openai/serving_embedding.py +9 -3
  57. sglang/srt/entrypoints/openai/serving_rerank.py +3 -1
  58. sglang/srt/entrypoints/openai/serving_responses.py +48 -3
  59. sglang/srt/entrypoints/openai/serving_score.py +1 -0
  60. sglang/srt/environ.py +285 -0
  61. sglang/srt/eplb/eplb_manager.py +2 -2
  62. sglang/srt/eplb/expert_distribution.py +26 -13
  63. sglang/srt/eplb/expert_location.py +38 -8
  64. sglang/srt/eplb/expert_location_updater.py +1 -1
  65. sglang/srt/function_call/base_format_detector.py +3 -6
  66. sglang/srt/function_call/ebnf_composer.py +11 -9
  67. sglang/srt/function_call/function_call_parser.py +9 -2
  68. sglang/srt/function_call/glm4_moe_detector.py +4 -4
  69. sglang/srt/function_call/gpt_oss_detector.py +23 -0
  70. sglang/srt/function_call/json_array_parser.py +63 -0
  71. sglang/srt/function_call/kimik2_detector.py +17 -4
  72. sglang/srt/function_call/qwen3_coder_detector.py +1 -1
  73. sglang/srt/function_call/utils.py +96 -5
  74. sglang/srt/grpc/__init__.py +1 -0
  75. sglang/srt/grpc/compile_proto.py +245 -0
  76. sglang/srt/grpc/sglang_scheduler_pb2.py +111 -0
  77. sglang/srt/grpc/sglang_scheduler_pb2.pyi +434 -0
  78. sglang/srt/grpc/sglang_scheduler_pb2_grpc.py +239 -0
  79. sglang/srt/layers/activation.py +143 -9
  80. sglang/srt/layers/attention/aiter_backend.py +14 -15
  81. sglang/srt/layers/attention/ascend_backend.py +115 -9
  82. sglang/srt/layers/attention/attention_registry.py +206 -0
  83. sglang/srt/layers/attention/base_attn_backend.py +12 -3
  84. sglang/srt/layers/attention/cutlass_mla_backend.py +3 -3
  85. sglang/srt/layers/attention/dual_chunk_flashattention_backend.py +1 -1
  86. sglang/srt/layers/attention/fla/chunk.py +242 -0
  87. sglang/srt/layers/attention/fla/chunk_delta_h.py +314 -0
  88. sglang/srt/layers/attention/fla/chunk_o.py +178 -0
  89. sglang/srt/layers/attention/fla/chunk_scaled_dot_kkt.py +151 -0
  90. sglang/srt/layers/attention/fla/cumsum.py +300 -0
  91. sglang/srt/layers/attention/fla/fused_recurrent.py +640 -0
  92. sglang/srt/layers/attention/fla/fused_sigmoid_gating_recurrent.py +232 -0
  93. sglang/srt/layers/attention/fla/index.py +37 -0
  94. sglang/srt/layers/attention/fla/l2norm.py +150 -0
  95. sglang/srt/layers/attention/fla/layernorm_gated.py +326 -0
  96. sglang/srt/layers/attention/fla/op.py +66 -0
  97. sglang/srt/layers/attention/fla/solve_tril.py +465 -0
  98. sglang/srt/layers/attention/fla/utils.py +331 -0
  99. sglang/srt/layers/attention/fla/wy_fast.py +158 -0
  100. sglang/srt/layers/attention/flashattention_backend.py +41 -8
  101. sglang/srt/layers/attention/flashinfer_backend.py +118 -198
  102. sglang/srt/layers/attention/flashinfer_mla_backend.py +27 -27
  103. sglang/srt/layers/attention/flashmla_backend.py +7 -5
  104. sglang/srt/layers/attention/hybrid_attn_backend.py +68 -53
  105. sglang/srt/layers/attention/hybrid_linear_attn_backend.py +602 -0
  106. sglang/srt/layers/attention/intel_amx_backend.py +3 -0
  107. sglang/srt/layers/attention/mamba/causal_conv1d.py +129 -0
  108. sglang/srt/layers/attention/mamba/causal_conv1d_triton.py +969 -0
  109. sglang/srt/layers/attention/mamba/mamba.py +629 -0
  110. sglang/srt/layers/attention/mamba/mamba_utils.py +81 -0
  111. sglang/srt/layers/attention/mamba/ops/__init__.py +2 -0
  112. sglang/srt/layers/attention/mamba/ops/layernorm_gated.py +172 -0
  113. sglang/srt/layers/attention/mamba/ops/mamba_ssm.py +442 -0
  114. sglang/srt/layers/attention/mamba/ops/ssd_bmm.py +264 -0
  115. sglang/srt/layers/attention/mamba/ops/ssd_chunk_scan.py +622 -0
  116. sglang/srt/layers/attention/mamba/ops/ssd_chunk_state.py +757 -0
  117. sglang/srt/layers/attention/mamba/ops/ssd_combined.py +262 -0
  118. sglang/srt/layers/attention/mamba/ops/ssd_state_passing.py +275 -0
  119. sglang/srt/layers/attention/npu_ops/mla_preprocess.py +393 -0
  120. sglang/srt/layers/attention/nsa/dequant_k_cache.py +163 -0
  121. sglang/srt/layers/attention/nsa/index_buf_accessor.py +354 -0
  122. sglang/srt/layers/attention/nsa/nsa_indexer.py +761 -0
  123. sglang/srt/layers/attention/nsa/quant_k_cache.py +255 -0
  124. sglang/srt/layers/attention/nsa/tilelang_kernel.py +785 -0
  125. sglang/srt/layers/attention/nsa/transform_index.py +144 -0
  126. sglang/srt/layers/attention/nsa/utils.py +24 -0
  127. sglang/srt/layers/attention/nsa_backend.py +887 -0
  128. sglang/srt/layers/attention/tbo_backend.py +6 -6
  129. sglang/srt/layers/attention/torch_flex_backend.py +325 -0
  130. sglang/srt/layers/attention/torch_native_backend.py +12 -6
  131. sglang/srt/layers/attention/triton_backend.py +57 -7
  132. sglang/srt/layers/attention/trtllm_mha_backend.py +5 -7
  133. sglang/srt/layers/attention/trtllm_mla_backend.py +276 -39
  134. sglang/srt/layers/attention/vision.py +58 -0
  135. sglang/srt/layers/attention/wave_backend.py +4 -4
  136. sglang/srt/layers/attention/wave_ops/decode_attention.py +2 -4
  137. sglang/srt/layers/attention/wave_ops/extend_attention.py +1 -3
  138. sglang/srt/layers/communicator.py +8 -0
  139. sglang/srt/layers/dp_attention.py +41 -2
  140. sglang/srt/layers/elementwise.py +3 -1
  141. sglang/srt/layers/layernorm.py +34 -15
  142. sglang/srt/layers/linear.py +55 -7
  143. sglang/srt/layers/logits_processor.py +44 -12
  144. sglang/srt/layers/moe/__init__.py +2 -1
  145. sglang/srt/layers/moe/cutlass_w4a8_moe.py +3 -3
  146. sglang/srt/layers/moe/ep_moe/kernels.py +2 -2
  147. sglang/srt/layers/moe/ep_moe/layer.py +256 -63
  148. sglang/srt/layers/moe/flashinfer_cutedsl_moe.py +183 -0
  149. sglang/srt/layers/moe/fused_moe_native.py +5 -3
  150. sglang/srt/layers/moe/fused_moe_triton/configs/{triton_3_4_0/E=128,N=768,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json → triton_3_3_1/E=128,N=768,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json } +35 -35
  151. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=128,N=352,device_name=NVIDIA_RTX_5880_Ada_Generation,dtype=fp8_w8a8.json +146 -0
  152. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
  153. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_H20.json +146 -0
  154. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  155. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H20-3e.json +146 -0
  156. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H200.json +146 -0
  157. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H800,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
  158. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_B200.json +146 -0
  159. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H20-3e.json +146 -0
  160. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H200.json +146 -0
  161. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H100_80GB_HBM3.json +146 -0
  162. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H200.json +146 -0
  163. sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +5 -2
  164. sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py +7 -3
  165. sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py +23 -20
  166. sglang/srt/layers/moe/fused_moe_triton/layer.py +71 -70
  167. sglang/srt/layers/moe/moe_runner/__init__.py +2 -1
  168. sglang/srt/layers/moe/moe_runner/base.py +274 -1
  169. sglang/srt/layers/moe/moe_runner/runner.py +80 -0
  170. sglang/srt/layers/moe/moe_runner/triton.py +448 -0
  171. sglang/srt/layers/moe/token_dispatcher/__init__.py +16 -4
  172. sglang/srt/layers/moe/token_dispatcher/{base_dispatcher.py → base.py} +67 -17
  173. sglang/srt/layers/moe/token_dispatcher/deepep.py +118 -56
  174. sglang/srt/layers/moe/token_dispatcher/standard.py +44 -2
  175. sglang/srt/layers/moe/topk.py +30 -9
  176. sglang/srt/layers/moe/utils.py +22 -6
  177. sglang/srt/layers/parameter.py +23 -6
  178. sglang/srt/layers/quantization/awq.py +19 -7
  179. sglang/srt/layers/quantization/base_config.py +11 -6
  180. sglang/srt/layers/quantization/blockwise_int8.py +38 -27
  181. sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py +1 -0
  182. sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py +50 -30
  183. sglang/srt/layers/quantization/compressed_tensors/schemes/__init__.py +2 -0
  184. sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py +13 -1
  185. sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py +173 -0
  186. sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py +2 -10
  187. sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py +27 -0
  188. sglang/srt/layers/quantization/fp8.py +78 -49
  189. sglang/srt/layers/quantization/fp8_utils.py +51 -32
  190. sglang/srt/layers/quantization/gptq.py +25 -17
  191. sglang/srt/layers/quantization/modelopt_quant.py +190 -55
  192. sglang/srt/layers/quantization/moe_wna16.py +21 -18
  193. sglang/srt/layers/quantization/mxfp4.py +74 -42
  194. sglang/srt/layers/quantization/quark/quark_moe.py +48 -30
  195. sglang/srt/layers/quantization/unquant.py +135 -47
  196. sglang/srt/layers/quantization/w4afp8.py +26 -17
  197. sglang/srt/layers/quantization/w8a8_fp8.py +35 -20
  198. sglang/srt/layers/quantization/w8a8_int8.py +91 -41
  199. sglang/srt/layers/rotary_embedding.py +78 -31
  200. sglang/srt/layers/sampler.py +213 -21
  201. sglang/srt/layers/utils.py +23 -0
  202. sglang/srt/lora/backend/base_backend.py +50 -8
  203. sglang/srt/lora/backend/chunked_backend.py +348 -0
  204. sglang/srt/lora/backend/triton_backend.py +99 -5
  205. sglang/srt/lora/layers.py +32 -0
  206. sglang/srt/lora/lora.py +8 -3
  207. sglang/srt/lora/lora_manager.py +52 -118
  208. sglang/srt/lora/mem_pool.py +25 -11
  209. sglang/srt/lora/triton_ops/__init__.py +4 -0
  210. sglang/srt/lora/triton_ops/chunked_sgmv_expand.py +214 -0
  211. sglang/srt/lora/triton_ops/chunked_sgmv_shrink.py +174 -0
  212. sglang/srt/lora/utils.py +22 -11
  213. sglang/srt/managers/async_dynamic_batch_tokenizer.py +170 -0
  214. sglang/srt/managers/cache_controller.py +199 -301
  215. sglang/srt/managers/data_parallel_controller.py +115 -80
  216. sglang/srt/managers/detokenizer_manager.py +19 -15
  217. sglang/srt/managers/disagg_service.py +46 -0
  218. sglang/srt/managers/io_struct.py +340 -109
  219. sglang/srt/managers/mm_utils.py +44 -6
  220. sglang/srt/managers/multi_tokenizer_mixin.py +357 -407
  221. sglang/srt/managers/multimodal_processor.py +1 -2
  222. sglang/srt/managers/overlap_utils.py +53 -0
  223. sglang/srt/managers/schedule_batch.py +240 -138
  224. sglang/srt/managers/schedule_policy.py +144 -17
  225. sglang/srt/managers/scheduler.py +502 -209
  226. sglang/srt/managers/scheduler_input_blocker.py +1 -1
  227. sglang/srt/managers/scheduler_metrics_mixin.py +99 -126
  228. sglang/srt/managers/scheduler_output_processor_mixin.py +75 -22
  229. sglang/srt/managers/scheduler_profiler_mixin.py +6 -6
  230. sglang/srt/managers/scheduler_update_weights_mixin.py +7 -0
  231. sglang/srt/managers/tokenizer_communicator_mixin.py +675 -0
  232. sglang/srt/managers/tokenizer_manager.py +320 -632
  233. sglang/srt/managers/tp_worker.py +81 -22
  234. sglang/srt/managers/tp_worker_overlap_thread.py +71 -56
  235. sglang/srt/managers/utils.py +1 -45
  236. sglang/srt/mem_cache/allocator.py +14 -20
  237. sglang/srt/mem_cache/allocator_ascend.py +41 -27
  238. sglang/srt/mem_cache/base_prefix_cache.py +1 -1
  239. sglang/srt/mem_cache/chunk_cache.py +8 -1
  240. sglang/srt/mem_cache/evict_policy.py +23 -0
  241. sglang/srt/mem_cache/hicache_storage.py +43 -24
  242. sglang/srt/mem_cache/hiradix_cache.py +222 -75
  243. sglang/srt/mem_cache/memory_pool.py +535 -58
  244. sglang/srt/mem_cache/memory_pool_host.py +239 -228
  245. sglang/srt/mem_cache/radix_cache.py +222 -73
  246. sglang/srt/mem_cache/radix_cache_cpp.py +11 -8
  247. sglang/srt/mem_cache/storage/__init__.py +10 -0
  248. sglang/srt/mem_cache/storage/aibrix_kvcache/aibrix_kvcache_storage.py +151 -0
  249. sglang/srt/mem_cache/storage/aibrix_kvcache/unit_test.py +109 -0
  250. sglang/srt/mem_cache/storage/backend_factory.py +223 -0
  251. sglang/srt/mem_cache/storage/eic/eic_storage.py +778 -0
  252. sglang/srt/mem_cache/storage/eic/test_unit.py +115 -0
  253. sglang/srt/mem_cache/storage/hf3fs/hf3fs_client.py +164 -0
  254. sglang/srt/mem_cache/storage/hf3fs/{client_hf3fs.py → hf3fs_usrbio_client.py} +5 -1
  255. sglang/srt/mem_cache/storage/hf3fs/storage_hf3fs.py +259 -62
  256. sglang/srt/mem_cache/storage/lmcache/lmc_radix_cache.py +284 -0
  257. sglang/srt/mem_cache/storage/lmcache/unit_test.py +121 -0
  258. sglang/srt/mem_cache/storage/mooncake_store/mooncake_store.py +166 -17
  259. sglang/srt/mem_cache/swa_radix_cache.py +25 -36
  260. sglang/srt/metrics/collector.py +511 -132
  261. sglang/srt/metrics/func_timer.py +2 -7
  262. sglang/srt/metrics/startup_func_log_and_timer.py +150 -0
  263. sglang/srt/metrics/utils.py +8 -1
  264. sglang/srt/model_executor/cpu_graph_runner.py +640 -0
  265. sglang/srt/model_executor/cuda_graph_runner.py +52 -37
  266. sglang/srt/model_executor/forward_batch_info.py +82 -40
  267. sglang/srt/model_executor/model_runner.py +432 -157
  268. sglang/srt/model_executor/npu_graph_runner.py +12 -5
  269. sglang/srt/model_loader/__init__.py +9 -3
  270. sglang/srt/model_loader/loader.py +133 -5
  271. sglang/srt/model_loader/remote_instance_weight_loader_utils.py +69 -0
  272. sglang/srt/model_loader/weight_utils.py +158 -3
  273. sglang/srt/models/apertus.py +686 -0
  274. sglang/srt/models/bailing_moe.py +820 -217
  275. sglang/srt/models/bailing_moe_nextn.py +168 -0
  276. sglang/srt/models/deepseek_nextn.py +6 -1
  277. sglang/srt/models/deepseek_v2.py +607 -130
  278. sglang/srt/models/dots_ocr.py +173 -0
  279. sglang/srt/models/dots_vlm.py +174 -0
  280. sglang/srt/models/dots_vlm_vit.py +337 -0
  281. sglang/srt/models/ernie4.py +1 -1
  282. sglang/srt/models/falcon_h1.py +576 -0
  283. sglang/srt/models/gemma3_causal.py +0 -2
  284. sglang/srt/models/gemma3_mm.py +1 -1
  285. sglang/srt/models/gemma3n_mm.py +2 -2
  286. sglang/srt/models/glm4_moe.py +4 -4
  287. sglang/srt/models/glm4_moe_nextn.py +2 -2
  288. sglang/srt/models/glm4v.py +5 -3
  289. sglang/srt/models/glm4v_moe.py +4 -1
  290. sglang/srt/models/gpt_oss.py +8 -31
  291. sglang/srt/models/kimi_vl_moonvit.py +2 -2
  292. sglang/srt/models/llama.py +4 -0
  293. sglang/srt/models/llama4.py +9 -0
  294. sglang/srt/models/llama_eagle3.py +13 -0
  295. sglang/srt/models/longcat_flash.py +3 -3
  296. sglang/srt/models/longcat_flash_nextn.py +1 -1
  297. sglang/srt/models/mllama4.py +40 -4
  298. sglang/srt/models/opt.py +637 -0
  299. sglang/srt/models/qwen2_5_vl.py +29 -5
  300. sglang/srt/models/qwen2_audio.py +1 -1
  301. sglang/srt/models/qwen2_moe.py +120 -13
  302. sglang/srt/models/qwen2_vl.py +1 -1
  303. sglang/srt/models/qwen3.py +18 -3
  304. sglang/srt/models/qwen3_moe.py +32 -4
  305. sglang/srt/models/qwen3_next.py +1069 -0
  306. sglang/srt/models/qwen3_next_mtp.py +112 -0
  307. sglang/srt/models/qwen3_vl.py +787 -0
  308. sglang/srt/models/qwen3_vl_moe.py +471 -0
  309. sglang/srt/models/registry.py +15 -3
  310. sglang/srt/models/sarashina2_vision.py +269 -0
  311. sglang/srt/models/solar.py +505 -0
  312. sglang/srt/models/starcoder2.py +357 -0
  313. sglang/srt/models/step3_vl.py +1 -1
  314. sglang/srt/models/torch_native_llama.py +9 -2
  315. sglang/srt/models/utils.py +51 -0
  316. sglang/srt/multimodal/processors/base_processor.py +15 -7
  317. sglang/srt/multimodal/processors/dots_vlm.py +98 -0
  318. sglang/srt/multimodal/processors/glm4v.py +9 -9
  319. sglang/srt/multimodal/processors/internvl.py +153 -129
  320. sglang/srt/multimodal/processors/qwen_vl.py +23 -6
  321. sglang/srt/multimodal/processors/sarashina2_vision.py +81 -0
  322. sglang/srt/offloader.py +27 -3
  323. sglang/srt/parser/jinja_template_utils.py +6 -0
  324. sglang/srt/sampling/sampling_batch_info.py +38 -17
  325. sglang/srt/sampling/sampling_params.py +7 -0
  326. sglang/srt/server_args.py +966 -267
  327. sglang/srt/server_args_config_parser.py +146 -0
  328. sglang/srt/single_batch_overlap.py +151 -0
  329. sglang/srt/speculative/cpp_ngram/ngram.cpp +374 -0
  330. sglang/srt/speculative/cpp_ngram/ngram.h +110 -0
  331. sglang/srt/speculative/cpp_ngram/ngram_cache.py +138 -0
  332. sglang/srt/speculative/cpp_ngram/ngram_cache_binding.cpp +43 -0
  333. sglang/srt/speculative/cpp_ngram/param.h +125 -0
  334. sglang/srt/speculative/cpp_ngram/queue.h +71 -0
  335. sglang/srt/speculative/eagle_draft_cuda_graph_runner.py +7 -1
  336. sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py +13 -2
  337. sglang/srt/speculative/{eagle_utils.py → eagle_info.py} +207 -757
  338. sglang/srt/speculative/eagle_worker.py +99 -28
  339. sglang/srt/speculative/ngram_utils.py +428 -0
  340. sglang/srt/speculative/ngram_worker.py +245 -0
  341. sglang/srt/speculative/spec_info.py +52 -0
  342. sglang/srt/speculative/spec_utils.py +606 -0
  343. sglang/srt/speculative/standalone_worker.py +109 -0
  344. sglang/srt/torch_memory_saver_adapter.py +5 -7
  345. sglang/srt/tracing/trace.py +578 -0
  346. sglang/srt/two_batch_overlap.py +8 -5
  347. sglang/srt/utils/__init__.py +2 -0
  348. sglang/srt/{utils.py → utils/common.py} +433 -77
  349. sglang/srt/{hf_transformers_utils.py → utils/hf_transformers_utils.py} +53 -5
  350. sglang/srt/{patch_torch.py → utils/patch_torch.py} +8 -0
  351. sglang/srt/utils/rpd_utils.py +452 -0
  352. sglang/srt/utils/slow_rank_detector.py +71 -0
  353. sglang/srt/warmup.py +8 -4
  354. sglang/srt/weight_sync/utils.py +2 -2
  355. sglang/test/attention/test_trtllm_mla_backend.py +169 -5
  356. sglang/test/get_logits_ut.py +57 -0
  357. sglang/test/run_eval.py +79 -11
  358. sglang/test/runners.py +5 -1
  359. sglang/test/simple_eval_common.py +5 -2
  360. sglang/test/simple_eval_mmmu_vlm.py +441 -0
  361. sglang/test/test_block_fp8.py +2 -2
  362. sglang/test/test_cutlass_moe.py +24 -6
  363. sglang/test/test_deterministic.py +297 -0
  364. sglang/test/test_disaggregation_utils.py +77 -0
  365. sglang/test/test_fp4_moe.py +370 -1
  366. sglang/test/test_programs.py +1 -1
  367. sglang/test/test_utils.py +383 -5
  368. sglang/utils.py +21 -1
  369. sglang/version.py +1 -1
  370. {sglang-0.5.2rc2.dist-info → sglang-0.5.3.dist-info}/METADATA +69 -124
  371. {sglang-0.5.2rc2.dist-info → sglang-0.5.3.dist-info}/RECORD +375 -245
  372. sglang/srt/disaggregation/launch_lb.py +0 -118
  373. sglang/srt/mem_cache/lora_radix_cache.py +0 -421
  374. /sglang/srt/{poll_based_barrier.py → utils/poll_based_barrier.py} +0 -0
  375. {sglang-0.5.2rc2.dist-info → sglang-0.5.3.dist-info}/WHEEL +0 -0
  376. {sglang-0.5.2rc2.dist-info → sglang-0.5.3.dist-info}/licenses/LICENSE +0 -0
  377. {sglang-0.5.2rc2.dist-info → sglang-0.5.3.dist-info}/top_level.txt +0 -0
@@ -1,7 +1,7 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: sglang
3
- Version: 0.5.2rc2
4
- Summary: SGLang is yet another fast serving framework for large language models and vision language models.
3
+ Version: 0.5.3
4
+ Summary: SGLang is a fast serving framework for large language models and vision language models.
5
5
  License: Apache License
6
6
  Version 2.0, January 2004
7
7
  http://www.apache.org/licenses/
@@ -211,88 +211,61 @@ Classifier: License :: OSI Approved :: Apache Software License
211
211
  Requires-Python: >=3.10
212
212
  Description-Content-Type: text/markdown
213
213
  License-File: LICENSE
214
+ Requires-Dist: IPython
214
215
  Requires-Dist: aiohttp
215
- Requires-Dist: requests
216
- Requires-Dist: tqdm
216
+ Requires-Dist: anthropic>=0.20.0
217
+ Requires-Dist: blobfile==3.0.0
218
+ Requires-Dist: build
219
+ Requires-Dist: compressed-tensors
220
+ Requires-Dist: cuda-python
221
+ Requires-Dist: datasets
222
+ Requires-Dist: einops
223
+ Requires-Dist: fastapi
224
+ Requires-Dist: flashinfer_python==0.4.0rc3
225
+ Requires-Dist: hf_transfer
226
+ Requires-Dist: huggingface_hub
227
+ Requires-Dist: interegular
228
+ Requires-Dist: llguidance<0.8.0,>=0.7.11
229
+ Requires-Dist: modelscope
230
+ Requires-Dist: msgspec
231
+ Requires-Dist: ninja
217
232
  Requires-Dist: numpy
218
- Requires-Dist: IPython
233
+ Requires-Dist: nvidia-cutlass-dsl==4.2.1
234
+ Requires-Dist: openai-harmony==0.0.4
235
+ Requires-Dist: openai==1.99.1
236
+ Requires-Dist: orjson
237
+ Requires-Dist: outlines==0.1.11
238
+ Requires-Dist: packaging
239
+ Requires-Dist: partial_json_parser
240
+ Requires-Dist: pillow
241
+ Requires-Dist: prometheus-client>=0.20.0
242
+ Requires-Dist: psutil
243
+ Requires-Dist: py-spy
244
+ Requires-Dist: pybase64
245
+ Requires-Dist: pydantic
246
+ Requires-Dist: pynvml
247
+ Requires-Dist: python-multipart
248
+ Requires-Dist: pyzmq>=25.1.2
249
+ Requires-Dist: requests
250
+ Requires-Dist: scipy
251
+ Requires-Dist: sentencepiece
219
252
  Requires-Dist: setproctitle
220
- Provides-Extra: runtime-common
221
- Requires-Dist: blobfile==3.0.0; extra == "runtime-common"
222
- Requires-Dist: build; extra == "runtime-common"
223
- Requires-Dist: compressed-tensors; extra == "runtime-common"
224
- Requires-Dist: datasets; extra == "runtime-common"
225
- Requires-Dist: einops; extra == "runtime-common"
226
- Requires-Dist: fastapi; extra == "runtime-common"
227
- Requires-Dist: hf_transfer; extra == "runtime-common"
228
- Requires-Dist: huggingface_hub; extra == "runtime-common"
229
- Requires-Dist: interegular; extra == "runtime-common"
230
- Requires-Dist: llguidance<0.8.0,>=0.7.11; extra == "runtime-common"
231
- Requires-Dist: modelscope; extra == "runtime-common"
232
- Requires-Dist: msgspec; extra == "runtime-common"
233
- Requires-Dist: ninja; extra == "runtime-common"
234
- Requires-Dist: openai==1.99.1; extra == "runtime-common"
235
- Requires-Dist: openai-harmony==0.0.4; extra == "runtime-common"
236
- Requires-Dist: orjson; extra == "runtime-common"
237
- Requires-Dist: outlines==0.1.11; extra == "runtime-common"
238
- Requires-Dist: packaging; extra == "runtime-common"
239
- Requires-Dist: partial_json_parser; extra == "runtime-common"
240
- Requires-Dist: pillow; extra == "runtime-common"
241
- Requires-Dist: prometheus-client>=0.20.0; extra == "runtime-common"
242
- Requires-Dist: psutil; extra == "runtime-common"
243
- Requires-Dist: pybase64; extra == "runtime-common"
244
- Requires-Dist: pydantic; extra == "runtime-common"
245
- Requires-Dist: pynvml; extra == "runtime-common"
246
- Requires-Dist: python-multipart; extra == "runtime-common"
247
- Requires-Dist: pyzmq>=25.1.2; extra == "runtime-common"
248
- Requires-Dist: sentencepiece; extra == "runtime-common"
249
- Requires-Dist: soundfile==0.13.1; extra == "runtime-common"
250
- Requires-Dist: scipy; extra == "runtime-common"
251
- Requires-Dist: timm==1.0.16; extra == "runtime-common"
252
- Requires-Dist: tiktoken; extra == "runtime-common"
253
- Requires-Dist: torchao==0.9.0; extra == "runtime-common"
254
- Requires-Dist: transformers==4.56.0; extra == "runtime-common"
255
- Requires-Dist: uvicorn; extra == "runtime-common"
256
- Requires-Dist: uvloop; extra == "runtime-common"
257
- Requires-Dist: xgrammar==0.1.23; extra == "runtime-common"
258
- Provides-Extra: srt
259
- Requires-Dist: sglang[runtime_common]; extra == "srt"
260
- Requires-Dist: sgl-kernel==0.3.8; extra == "srt"
261
- Requires-Dist: torch==2.8.0; extra == "srt"
262
- Requires-Dist: torchaudio==2.8.0; extra == "srt"
263
- Requires-Dist: torchvision; extra == "srt"
264
- Requires-Dist: cuda-python; extra == "srt"
265
- Requires-Dist: flashinfer_python==0.3.0; extra == "srt"
266
- Provides-Extra: blackwell
267
- Requires-Dist: sglang[runtime_common]; extra == "blackwell"
268
- Requires-Dist: sgl-kernel; extra == "blackwell"
269
- Requires-Dist: torch==2.8.0; extra == "blackwell"
270
- Requires-Dist: torchaudio==2.8.0; extra == "blackwell"
271
- Requires-Dist: torchvision; extra == "blackwell"
272
- Requires-Dist: cuda-python; extra == "blackwell"
273
- Requires-Dist: flashinfer_python==0.3.0; extra == "blackwell"
274
- Provides-Extra: srt-hip
275
- Requires-Dist: sglang[runtime_common]; extra == "srt-hip"
276
- Requires-Dist: torch; extra == "srt-hip"
277
- Requires-Dist: petit_kernel==0.0.2; extra == "srt-hip"
278
- Requires-Dist: wave-lang==1.0.1; extra == "srt-hip"
279
- Provides-Extra: srt-cpu
280
- Requires-Dist: sglang[runtime_common]; extra == "srt-cpu"
281
- Provides-Extra: srt-npu
282
- Requires-Dist: sglang[runtime_common]; extra == "srt-npu"
283
- Provides-Extra: srt-xpu
284
- Requires-Dist: sglang[runtime_common]; extra == "srt-xpu"
285
- Provides-Extra: srt-hpu
286
- Requires-Dist: sglang[runtime_common]; extra == "srt-hpu"
287
- Provides-Extra: openai
288
- Requires-Dist: openai==1.99.1; extra == "openai"
289
- Requires-Dist: tiktoken; extra == "openai"
290
- Provides-Extra: anthropic
291
- Requires-Dist: anthropic>=0.20.0; extra == "anthropic"
292
- Provides-Extra: litellm
293
- Requires-Dist: litellm>=1.0.0; extra == "litellm"
294
- Provides-Extra: torch-memory-saver
295
- Requires-Dist: torch_memory_saver==0.0.8; extra == "torch-memory-saver"
253
+ Requires-Dist: sgl-kernel==0.3.14.post1
254
+ Requires-Dist: soundfile==0.13.1
255
+ Requires-Dist: tiktoken
256
+ Requires-Dist: timm==1.0.16
257
+ Requires-Dist: torch==2.8.0
258
+ Requires-Dist: torch_memory_saver==0.0.9rc2
259
+ Requires-Dist: torchao==0.9.0
260
+ Requires-Dist: torchaudio==2.8.0
261
+ Requires-Dist: torchvision
262
+ Requires-Dist: tqdm
263
+ Requires-Dist: transformers==4.57.0
264
+ Requires-Dist: uvicorn
265
+ Requires-Dist: uvloop
266
+ Requires-Dist: xgrammar==0.1.24
267
+ Requires-Dist: grpcio==1.75.1
268
+ Requires-Dist: grpcio-tools==1.75.1
296
269
  Provides-Extra: decord
297
270
  Requires-Dist: decord; extra == "decord"
298
271
  Provides-Extra: test
@@ -302,55 +275,27 @@ Requires-Dist: jsonlines; extra == "test"
302
275
  Requires-Dist: matplotlib; extra == "test"
303
276
  Requires-Dist: pandas; extra == "test"
304
277
  Requires-Dist: peft; extra == "test"
305
- Requires-Dist: sentence_transformers; extra == "test"
306
278
  Requires-Dist: pytest; extra == "test"
279
+ Requires-Dist: sentence_transformers; extra == "test"
307
280
  Requires-Dist: tabulate; extra == "test"
281
+ Provides-Extra: tracing
282
+ Requires-Dist: opentelemetry-api; extra == "tracing"
283
+ Requires-Dist: opentelemetry-exporter-otlp; extra == "tracing"
284
+ Requires-Dist: opentelemetry-exporter-otlp-proto-grpc; extra == "tracing"
285
+ Requires-Dist: opentelemetry-sdk; extra == "tracing"
308
286
  Provides-Extra: all
309
- Requires-Dist: sglang[srt]; extra == "all"
310
- Requires-Dist: sglang[openai]; extra == "all"
311
- Requires-Dist: sglang[anthropic]; extra == "all"
312
- Requires-Dist: sglang[torch_memory_saver]; extra == "all"
287
+ Requires-Dist: sglang[test]; extra == "all"
313
288
  Requires-Dist: sglang[decord]; extra == "all"
314
- Provides-Extra: all-hip
315
- Requires-Dist: sglang[srt_hip]; extra == "all-hip"
316
- Requires-Dist: sglang[openai]; extra == "all-hip"
317
- Requires-Dist: sglang[anthropic]; extra == "all-hip"
318
- Requires-Dist: sglang[decord]; extra == "all-hip"
319
- Provides-Extra: all-xpu
320
- Requires-Dist: sglang[srt_xpu]; extra == "all-xpu"
321
- Requires-Dist: sglang[openai]; extra == "all-xpu"
322
- Requires-Dist: sglang[anthropic]; extra == "all-xpu"
323
- Requires-Dist: sglang[decord]; extra == "all-xpu"
324
- Provides-Extra: all-hpu
325
- Requires-Dist: sglang[srt_hpu]; extra == "all-hpu"
326
- Requires-Dist: sglang[openai]; extra == "all-hpu"
327
- Requires-Dist: sglang[anthropic]; extra == "all-hpu"
328
- Requires-Dist: sglang[decord]; extra == "all-hpu"
329
- Provides-Extra: all-cpu
330
- Requires-Dist: sglang[srt_cpu]; extra == "all-cpu"
331
- Requires-Dist: sglang[openai]; extra == "all-cpu"
332
- Requires-Dist: sglang[anthropic]; extra == "all-cpu"
333
- Requires-Dist: sglang[decord]; extra == "all-cpu"
334
- Provides-Extra: all-npu
335
- Requires-Dist: sglang[srt_npu]; extra == "all-npu"
336
- Requires-Dist: sglang[openai]; extra == "all-npu"
337
- Requires-Dist: sglang[anthropic]; extra == "all-npu"
338
- Requires-Dist: sglang[decord]; extra == "all-npu"
289
+ Provides-Extra: all-aarch64
290
+ Requires-Dist: sglang[test]; extra == "all-aarch64"
339
291
  Provides-Extra: dev
340
- Requires-Dist: sglang[all]; extra == "dev"
341
292
  Requires-Dist: sglang[test]; extra == "dev"
342
- Provides-Extra: dev-hip
343
- Requires-Dist: sglang[all_hip]; extra == "dev-hip"
344
- Requires-Dist: sglang[test]; extra == "dev-hip"
345
- Provides-Extra: dev-xpu
346
- Requires-Dist: sglang[all_xpu]; extra == "dev-xpu"
347
- Requires-Dist: sglang[test]; extra == "dev-xpu"
348
- Provides-Extra: dev-hpu
349
- Requires-Dist: sglang[all_hpu]; extra == "dev-hpu"
350
- Requires-Dist: sglang[test]; extra == "dev-hpu"
351
- Provides-Extra: dev-cpu
352
- Requires-Dist: sglang[all_cpu]; extra == "dev-cpu"
353
- Requires-Dist: sglang[test]; extra == "dev-cpu"
293
+ Requires-Dist: sglang[decord]; extra == "dev"
294
+ Provides-Extra: blackwell
295
+ Requires-Dist: sglang[test]; extra == "blackwell"
296
+ Requires-Dist: sglang[decord]; extra == "blackwell"
297
+ Provides-Extra: blackwell-aarch64
298
+ Requires-Dist: sglang[test]; extra == "blackwell-aarch64"
354
299
  Dynamic: license-file
355
300
 
356
301
  <div align="center" id="sglangtop">