sglang 0.4.7__py3-none-any.whl → 0.4.8__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 (152) hide show
  1. sglang/__init__.py +2 -0
  2. sglang/api.py +7 -0
  3. sglang/bench_one_batch.py +8 -6
  4. sglang/bench_serving.py +1 -1
  5. sglang/lang/interpreter.py +40 -1
  6. sglang/lang/ir.py +27 -0
  7. sglang/math_utils.py +8 -0
  8. sglang/srt/_custom_ops.py +2 -2
  9. sglang/srt/code_completion_parser.py +2 -44
  10. sglang/srt/configs/model_config.py +6 -0
  11. sglang/srt/constants.py +3 -0
  12. sglang/srt/conversation.py +19 -3
  13. sglang/srt/custom_op.py +5 -1
  14. sglang/srt/disaggregation/base/__init__.py +1 -1
  15. sglang/srt/disaggregation/base/conn.py +25 -11
  16. sglang/srt/disaggregation/common/__init__.py +5 -1
  17. sglang/srt/disaggregation/common/utils.py +42 -0
  18. sglang/srt/disaggregation/decode.py +211 -72
  19. sglang/srt/disaggregation/decode_schedule_batch_mixin.py +4 -3
  20. sglang/srt/disaggregation/fake/__init__.py +1 -1
  21. sglang/srt/disaggregation/fake/conn.py +15 -9
  22. sglang/srt/disaggregation/mini_lb.py +34 -4
  23. sglang/srt/disaggregation/mooncake/__init__.py +1 -1
  24. sglang/srt/disaggregation/mooncake/conn.py +30 -29
  25. sglang/srt/disaggregation/nixl/__init__.py +6 -1
  26. sglang/srt/disaggregation/nixl/conn.py +17 -12
  27. sglang/srt/disaggregation/prefill.py +144 -55
  28. sglang/srt/disaggregation/utils.py +155 -123
  29. sglang/srt/distributed/parallel_state.py +12 -4
  30. sglang/srt/entrypoints/engine.py +37 -29
  31. sglang/srt/entrypoints/http_server.py +153 -72
  32. sglang/srt/entrypoints/http_server_engine.py +0 -3
  33. sglang/srt/entrypoints/openai/__init__.py +0 -0
  34. sglang/srt/{openai_api → entrypoints/openai}/protocol.py +84 -10
  35. sglang/srt/entrypoints/openai/serving_base.py +149 -0
  36. sglang/srt/entrypoints/openai/serving_chat.py +921 -0
  37. sglang/srt/entrypoints/openai/serving_completions.py +424 -0
  38. sglang/srt/entrypoints/openai/serving_embedding.py +169 -0
  39. sglang/srt/entrypoints/openai/serving_rerank.py +102 -0
  40. sglang/srt/entrypoints/openai/serving_score.py +61 -0
  41. sglang/srt/entrypoints/openai/usage_processor.py +81 -0
  42. sglang/srt/entrypoints/openai/utils.py +72 -0
  43. sglang/srt/eplb_simulator/__init__.py +1 -0
  44. sglang/srt/eplb_simulator/reader.py +51 -0
  45. sglang/srt/function_call/base_format_detector.py +7 -4
  46. sglang/srt/function_call/deepseekv3_detector.py +1 -1
  47. sglang/srt/function_call/ebnf_composer.py +64 -10
  48. sglang/srt/function_call/function_call_parser.py +6 -6
  49. sglang/srt/function_call/llama32_detector.py +1 -1
  50. sglang/srt/function_call/mistral_detector.py +1 -1
  51. sglang/srt/function_call/pythonic_detector.py +1 -1
  52. sglang/srt/function_call/qwen25_detector.py +1 -1
  53. sglang/srt/{openai_api/utils.py → jinja_template_utils.py} +6 -5
  54. sglang/srt/layers/activation.py +40 -3
  55. sglang/srt/layers/attention/aiter_backend.py +20 -4
  56. sglang/srt/layers/attention/base_attn_backend.py +1 -1
  57. sglang/srt/layers/attention/cutlass_mla_backend.py +39 -15
  58. sglang/srt/layers/attention/flashattention_backend.py +71 -72
  59. sglang/srt/layers/attention/flashinfer_backend.py +10 -8
  60. sglang/srt/layers/attention/flashinfer_mla_backend.py +29 -28
  61. sglang/srt/layers/attention/flashmla_backend.py +7 -12
  62. sglang/srt/layers/attention/tbo_backend.py +3 -3
  63. sglang/srt/layers/attention/triton_backend.py +138 -130
  64. sglang/srt/layers/attention/triton_ops/decode_attention.py +2 -7
  65. sglang/srt/layers/attention/vision.py +51 -24
  66. sglang/srt/layers/communicator.py +28 -10
  67. sglang/srt/layers/dp_attention.py +11 -2
  68. sglang/srt/layers/layernorm.py +29 -2
  69. sglang/srt/layers/linear.py +0 -4
  70. sglang/srt/layers/logits_processor.py +2 -14
  71. sglang/srt/layers/moe/ep_moe/kernels.py +165 -7
  72. sglang/srt/layers/moe/ep_moe/layer.py +249 -33
  73. sglang/srt/layers/moe/ep_moe/token_dispatcher.py +11 -37
  74. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=128,N=384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
  75. sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +7 -4
  76. sglang/srt/layers/moe/fused_moe_triton/layer.py +75 -12
  77. sglang/srt/layers/moe/topk.py +107 -12
  78. sglang/srt/layers/pooler.py +56 -0
  79. sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py +6 -2
  80. sglang/srt/layers/quantization/deep_gemm_wrapper/__init__.py +1 -0
  81. sglang/srt/layers/quantization/{deep_gemm.py → deep_gemm_wrapper/compile_utils.py} +23 -80
  82. sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py +32 -0
  83. sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py +110 -0
  84. sglang/srt/layers/quantization/fp8.py +25 -17
  85. sglang/srt/layers/quantization/fp8_kernel.py +44 -15
  86. sglang/srt/layers/quantization/fp8_utils.py +87 -22
  87. sglang/srt/layers/quantization/modelopt_quant.py +62 -8
  88. sglang/srt/layers/quantization/utils.py +5 -2
  89. sglang/srt/layers/radix_attention.py +2 -3
  90. sglang/srt/layers/rotary_embedding.py +42 -2
  91. sglang/srt/layers/sampler.py +1 -1
  92. sglang/srt/lora/lora_manager.py +249 -105
  93. sglang/srt/lora/mem_pool.py +53 -50
  94. sglang/srt/lora/utils.py +1 -1
  95. sglang/srt/managers/cache_controller.py +33 -14
  96. sglang/srt/managers/io_struct.py +31 -10
  97. sglang/srt/managers/multimodal_processors/base_processor.py +2 -2
  98. sglang/srt/managers/multimodal_processors/vila.py +85 -0
  99. sglang/srt/managers/schedule_batch.py +79 -37
  100. sglang/srt/managers/schedule_policy.py +70 -56
  101. sglang/srt/managers/scheduler.py +220 -79
  102. sglang/srt/managers/template_manager.py +226 -0
  103. sglang/srt/managers/tokenizer_manager.py +40 -10
  104. sglang/srt/managers/tp_worker.py +12 -2
  105. sglang/srt/managers/tp_worker_overlap_thread.py +11 -0
  106. sglang/srt/mem_cache/{paged_allocator.py → allocator.py} +125 -34
  107. sglang/srt/mem_cache/base_prefix_cache.py +52 -8
  108. sglang/srt/mem_cache/chunk_cache.py +11 -15
  109. sglang/srt/mem_cache/hiradix_cache.py +38 -25
  110. sglang/srt/mem_cache/memory_pool.py +213 -505
  111. sglang/srt/mem_cache/memory_pool_host.py +380 -0
  112. sglang/srt/mem_cache/radix_cache.py +56 -28
  113. sglang/srt/model_executor/cuda_graph_runner.py +198 -100
  114. sglang/srt/model_executor/forward_batch_info.py +32 -10
  115. sglang/srt/model_executor/model_runner.py +28 -12
  116. sglang/srt/model_loader/loader.py +16 -2
  117. sglang/srt/model_loader/weight_utils.py +11 -2
  118. sglang/srt/models/bert.py +113 -13
  119. sglang/srt/models/deepseek_nextn.py +29 -27
  120. sglang/srt/models/deepseek_v2.py +213 -173
  121. sglang/srt/models/glm4.py +312 -0
  122. sglang/srt/models/internvl.py +46 -102
  123. sglang/srt/models/mimo_mtp.py +2 -18
  124. sglang/srt/models/roberta.py +117 -9
  125. sglang/srt/models/vila.py +305 -0
  126. sglang/srt/reasoning_parser.py +21 -11
  127. sglang/srt/sampling/sampling_batch_info.py +24 -0
  128. sglang/srt/sampling/sampling_params.py +2 -0
  129. sglang/srt/server_args.py +351 -238
  130. sglang/srt/speculative/build_eagle_tree.py +1 -1
  131. sglang/srt/speculative/eagle_draft_cuda_graph_runner.py +131 -9
  132. sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py +130 -14
  133. sglang/srt/speculative/eagle_utils.py +468 -116
  134. sglang/srt/speculative/eagle_worker.py +258 -84
  135. sglang/srt/torch_memory_saver_adapter.py +19 -15
  136. sglang/srt/two_batch_overlap.py +4 -2
  137. sglang/srt/utils.py +235 -11
  138. sglang/test/attention/test_prefix_chunk_info.py +2 -0
  139. sglang/test/runners.py +38 -3
  140. sglang/test/test_block_fp8.py +1 -0
  141. sglang/test/test_block_fp8_deep_gemm_blackwell.py +252 -0
  142. sglang/test/test_block_fp8_ep.py +2 -0
  143. sglang/test/test_utils.py +4 -1
  144. sglang/utils.py +9 -0
  145. sglang/version.py +1 -1
  146. {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/METADATA +8 -14
  147. {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/RECORD +150 -128
  148. sglang/srt/entrypoints/verl_engine.py +0 -179
  149. sglang/srt/openai_api/adapter.py +0 -1990
  150. {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/WHEEL +0 -0
  151. {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/licenses/LICENSE +0 -0
  152. {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/top_level.txt +0 -0
File without changes