sglang 0.4.10.post1__py3-none-any.whl → 0.5.0rc0__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 (143) hide show
  1. sglang/bench_one_batch.py +113 -17
  2. sglang/compile_deep_gemm.py +8 -1
  3. sglang/global_config.py +5 -1
  4. sglang/srt/configs/model_config.py +35 -0
  5. sglang/srt/conversation.py +9 -117
  6. sglang/srt/disaggregation/base/conn.py +5 -2
  7. sglang/srt/disaggregation/decode.py +6 -1
  8. sglang/srt/disaggregation/decode_schedule_batch_mixin.py +4 -0
  9. sglang/srt/disaggregation/mooncake/conn.py +243 -135
  10. sglang/srt/disaggregation/prefill.py +3 -0
  11. sglang/srt/distributed/device_communicators/pynccl.py +7 -0
  12. sglang/srt/distributed/device_communicators/pynccl_allocator.py +133 -0
  13. sglang/srt/distributed/device_communicators/pynccl_wrapper.py +42 -3
  14. sglang/srt/distributed/parallel_state.py +22 -9
  15. sglang/srt/entrypoints/context.py +244 -0
  16. sglang/srt/entrypoints/engine.py +8 -5
  17. sglang/srt/entrypoints/harmony_utils.py +370 -0
  18. sglang/srt/entrypoints/http_server.py +106 -15
  19. sglang/srt/entrypoints/openai/protocol.py +227 -1
  20. sglang/srt/entrypoints/openai/serving_chat.py +278 -42
  21. sglang/srt/entrypoints/openai/serving_responses.py +1273 -0
  22. sglang/srt/entrypoints/openai/tool_server.py +174 -0
  23. sglang/srt/entrypoints/tool.py +87 -0
  24. sglang/srt/eplb/expert_distribution.py +4 -2
  25. sglang/srt/eplb/expert_location.py +5 -1
  26. sglang/srt/function_call/harmony_tool_parser.py +130 -0
  27. sglang/srt/hf_transformers_utils.py +55 -13
  28. sglang/srt/jinja_template_utils.py +8 -1
  29. sglang/srt/layers/attention/aiter_backend.py +5 -8
  30. sglang/srt/layers/attention/cutlass_mla_backend.py +3 -3
  31. sglang/srt/layers/attention/dual_chunk_flashattention_backend.py +1700 -0
  32. sglang/srt/layers/attention/flashattention_backend.py +7 -11
  33. sglang/srt/layers/attention/triton_backend.py +85 -14
  34. sglang/srt/layers/attention/triton_ops/decode_attention.py +17 -0
  35. sglang/srt/layers/attention/triton_ops/extend_attention.py +143 -98
  36. sglang/srt/layers/attention/trtllm_mha_backend.py +332 -0
  37. sglang/srt/layers/attention/trtllm_mla_backend.py +6 -6
  38. sglang/srt/layers/attention/vision.py +40 -15
  39. sglang/srt/layers/communicator.py +35 -8
  40. sglang/srt/layers/dp_attention.py +12 -0
  41. sglang/srt/layers/linear.py +9 -8
  42. sglang/srt/layers/logits_processor.py +9 -1
  43. sglang/srt/layers/moe/cutlass_moe.py +20 -6
  44. sglang/srt/layers/moe/ep_moe/layer.py +87 -107
  45. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=352,device_name=NVIDIA_RTX_6000_Ada_Generation,dtype=fp8_w8a8.json +146 -0
  46. sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +101 -12
  47. sglang/srt/layers/moe/fused_moe_triton/layer.py +442 -58
  48. sglang/srt/layers/moe/fused_moe_triton/triton_kernels_moe.py +169 -15
  49. sglang/srt/layers/moe/token_dispatcher/__init__.py +23 -0
  50. sglang/srt/layers/moe/token_dispatcher/base_dispatcher.py +12 -1
  51. sglang/srt/layers/moe/{ep_moe/token_dispatcher.py → token_dispatcher/deepep.py} +8 -15
  52. sglang/srt/layers/moe/topk.py +12 -3
  53. sglang/srt/layers/moe/utils.py +59 -0
  54. sglang/srt/layers/quantization/__init__.py +22 -0
  55. sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py +3 -2
  56. sglang/srt/layers/quantization/deep_gemm_wrapper/compile_utils.py +1 -1
  57. sglang/srt/layers/quantization/fp4.py +557 -0
  58. sglang/srt/layers/quantization/fp8.py +8 -7
  59. sglang/srt/layers/quantization/fp8_kernel.py +0 -4
  60. sglang/srt/layers/quantization/fp8_utils.py +29 -0
  61. sglang/srt/layers/quantization/modelopt_quant.py +259 -64
  62. sglang/srt/layers/quantization/mxfp4.py +651 -0
  63. sglang/srt/layers/quantization/mxfp4_tensor.py +133 -0
  64. sglang/srt/layers/quantization/quark/__init__.py +0 -0
  65. sglang/srt/layers/quantization/quark/schemes/__init__.py +6 -0
  66. sglang/srt/layers/quantization/quark/schemes/quark_scheme.py +55 -0
  67. sglang/srt/layers/quantization/quark/schemes/quark_w4a4_mxfp4.py +118 -0
  68. sglang/srt/layers/quantization/quark/utils.py +107 -0
  69. sglang/srt/layers/quantization/unquant.py +60 -6
  70. sglang/srt/layers/quantization/w4afp8.py +1 -1
  71. sglang/srt/layers/rotary_embedding.py +225 -1
  72. sglang/srt/layers/utils.py +9 -0
  73. sglang/srt/layers/vocab_parallel_embedding.py +15 -4
  74. sglang/srt/lora/lora_manager.py +70 -14
  75. sglang/srt/lora/lora_registry.py +10 -2
  76. sglang/srt/lora/mem_pool.py +43 -5
  77. sglang/srt/managers/cache_controller.py +61 -32
  78. sglang/srt/managers/data_parallel_controller.py +52 -2
  79. sglang/srt/managers/detokenizer_manager.py +1 -1
  80. sglang/srt/managers/io_struct.py +21 -4
  81. sglang/srt/managers/mm_utils.py +5 -11
  82. sglang/srt/managers/schedule_batch.py +30 -8
  83. sglang/srt/managers/schedule_policy.py +3 -1
  84. sglang/srt/managers/scheduler.py +170 -18
  85. sglang/srt/managers/scheduler_output_processor_mixin.py +1 -2
  86. sglang/srt/managers/scheduler_recv_skipper.py +37 -0
  87. sglang/srt/managers/scheduler_update_weights_mixin.py +6 -0
  88. sglang/srt/managers/template_manager.py +59 -22
  89. sglang/srt/managers/tokenizer_manager.py +137 -67
  90. sglang/srt/managers/tp_worker.py +3 -0
  91. sglang/srt/managers/tp_worker_overlap_thread.py +3 -0
  92. sglang/srt/managers/utils.py +45 -1
  93. sglang/srt/mem_cache/cpp_radix_tree/radix_tree.py +182 -0
  94. sglang/srt/mem_cache/hicache_storage.py +13 -21
  95. sglang/srt/mem_cache/hiradix_cache.py +53 -5
  96. sglang/srt/mem_cache/memory_pool_host.py +1 -1
  97. sglang/srt/mem_cache/multimodal_cache.py +33 -13
  98. sglang/srt/mem_cache/radix_cache_cpp.py +229 -0
  99. sglang/srt/mem_cache/storage/hf3fs/client_hf3fs.py +2 -2
  100. sglang/srt/mem_cache/storage/hf3fs/hf3fs_utils.cpp +35 -0
  101. sglang/srt/model_executor/cuda_graph_runner.py +24 -9
  102. sglang/srt/model_executor/forward_batch_info.py +48 -17
  103. sglang/srt/model_executor/model_runner.py +24 -2
  104. sglang/srt/model_loader/weight_utils.py +10 -0
  105. sglang/srt/models/bailing_moe.py +425 -0
  106. sglang/srt/models/deepseek_v2.py +95 -50
  107. sglang/srt/models/ernie4.py +426 -0
  108. sglang/srt/models/ernie4_eagle.py +203 -0
  109. sglang/srt/models/gemma3n_mm.py +39 -0
  110. sglang/srt/models/glm4_moe.py +102 -27
  111. sglang/srt/models/gpt_oss.py +1134 -0
  112. sglang/srt/models/grok.py +3 -3
  113. sglang/srt/models/llama4.py +13 -2
  114. sglang/srt/models/mixtral.py +3 -3
  115. sglang/srt/models/mllama4.py +428 -19
  116. sglang/srt/models/qwen2.py +6 -0
  117. sglang/srt/models/qwen2_moe.py +7 -4
  118. sglang/srt/models/qwen3_moe.py +39 -14
  119. sglang/srt/models/step3_vl.py +10 -1
  120. sglang/srt/models/transformers.py +2 -5
  121. sglang/srt/multimodal/processors/base_processor.py +4 -3
  122. sglang/srt/multimodal/processors/gemma3n.py +0 -7
  123. sglang/srt/multimodal/processors/step3_vl.py +3 -1
  124. sglang/srt/operations_strategy.py +1 -1
  125. sglang/srt/reasoning_parser.py +18 -39
  126. sglang/srt/server_args.py +218 -23
  127. sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py +18 -0
  128. sglang/srt/two_batch_overlap.py +163 -9
  129. sglang/srt/utils.py +41 -26
  130. sglang/srt/weight_sync/utils.py +1 -1
  131. sglang/test/runners.py +4 -4
  132. sglang/test/test_utils.py +4 -4
  133. sglang/version.py +1 -1
  134. {sglang-0.4.10.post1.dist-info → sglang-0.5.0rc0.dist-info}/METADATA +18 -15
  135. {sglang-0.4.10.post1.dist-info → sglang-0.5.0rc0.dist-info}/RECORD +143 -116
  136. /sglang/srt/mem_cache/{mooncake_store → storage/mooncake_store}/mooncake_store.py +0 -0
  137. /sglang/srt/mem_cache/{mooncake_store → storage/mooncake_store}/unit_test.py +0 -0
  138. /sglang/srt/mem_cache/{nixl → storage/nixl}/hicache_nixl.py +0 -0
  139. /sglang/srt/mem_cache/{nixl → storage/nixl}/nixl_utils.py +0 -0
  140. /sglang/srt/mem_cache/{nixl → storage/nixl}/test_hicache_nixl_storage.py +0 -0
  141. {sglang-0.4.10.post1.dist-info → sglang-0.5.0rc0.dist-info}/WHEEL +0 -0
  142. {sglang-0.4.10.post1.dist-info → sglang-0.5.0rc0.dist-info}/licenses/LICENSE +0 -0
  143. {sglang-0.4.10.post1.dist-info → sglang-0.5.0rc0.dist-info}/top_level.txt +0 -0