sglang 0.5.4.post1__py3-none-any.whl → 0.5.4.post2__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 (150) hide show
  1. sglang/bench_one_batch.py +149 -34
  2. sglang/bench_serving.py +18 -3
  3. sglang/compile_deep_gemm.py +13 -7
  4. sglang/srt/batch_invariant_ops/__init__.py +2 -0
  5. sglang/srt/batch_invariant_ops/batch_invariant_ops.py +120 -0
  6. sglang/srt/checkpoint_engine/__init__.py +9 -0
  7. sglang/srt/checkpoint_engine/update.py +317 -0
  8. sglang/srt/configs/__init__.py +2 -0
  9. sglang/srt/configs/deepseek_ocr.py +542 -10
  10. sglang/srt/configs/deepseekvl2.py +95 -194
  11. sglang/srt/configs/kimi_linear.py +160 -0
  12. sglang/srt/configs/mamba_utils.py +66 -0
  13. sglang/srt/configs/model_config.py +25 -2
  14. sglang/srt/constants.py +7 -0
  15. sglang/srt/debug_utils/tensor_dump_forward_hook.py +149 -0
  16. sglang/srt/disaggregation/decode.py +34 -6
  17. sglang/srt/disaggregation/nixl/conn.py +2 -2
  18. sglang/srt/disaggregation/prefill.py +25 -3
  19. sglang/srt/distributed/device_communicators/custom_all_reduce.py +3 -1
  20. sglang/srt/distributed/parallel_state.py +9 -5
  21. sglang/srt/entrypoints/engine.py +13 -5
  22. sglang/srt/entrypoints/http_server.py +22 -3
  23. sglang/srt/entrypoints/openai/protocol.py +7 -1
  24. sglang/srt/entrypoints/openai/serving_chat.py +42 -0
  25. sglang/srt/entrypoints/openai/serving_completions.py +10 -0
  26. sglang/srt/entrypoints/openai/serving_embedding.py +1 -0
  27. sglang/srt/environ.py +7 -0
  28. sglang/srt/eplb/expert_distribution.py +34 -1
  29. sglang/srt/eplb/expert_location.py +106 -36
  30. sglang/srt/grpc/compile_proto.py +3 -0
  31. sglang/srt/layers/attention/ascend_backend.py +233 -5
  32. sglang/srt/layers/attention/attention_registry.py +3 -0
  33. sglang/srt/layers/attention/fla/chunk_delta_h.py +61 -32
  34. sglang/srt/layers/attention/fla/fused_recurrent.py +17 -4
  35. sglang/srt/layers/attention/fla/kda.py +1359 -0
  36. sglang/srt/layers/attention/fla/layernorm_gated.py +7 -1
  37. sglang/srt/layers/attention/flashattention_backend.py +7 -6
  38. sglang/srt/layers/attention/flashinfer_mla_backend.py +3 -1
  39. sglang/srt/layers/attention/flashmla_backend.py +1 -1
  40. sglang/srt/layers/attention/hybrid_linear_attn_backend.py +223 -0
  41. sglang/srt/layers/attention/mamba/mamba.py +20 -11
  42. sglang/srt/layers/attention/nsa/dequant_k_cache.py +138 -6
  43. sglang/srt/layers/attention/nsa/nsa_indexer.py +45 -22
  44. sglang/srt/layers/attention/nsa/quant_k_cache.py +44 -12
  45. sglang/srt/layers/attention/nsa/transform_index.py +1 -1
  46. sglang/srt/layers/attention/nsa_backend.py +157 -23
  47. sglang/srt/layers/attention/triton_backend.py +4 -1
  48. sglang/srt/layers/attention/trtllm_mha_backend.py +10 -4
  49. sglang/srt/layers/attention/trtllm_mla_backend.py +10 -2
  50. sglang/srt/layers/communicator.py +23 -1
  51. sglang/srt/layers/layernorm.py +16 -2
  52. sglang/srt/layers/logits_processor.py +4 -20
  53. sglang/srt/layers/moe/ep_moe/layer.py +0 -18
  54. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json +146 -0
  55. sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=256,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128]_down.json +164 -0
  56. sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +68 -22
  57. sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py +43 -3
  58. sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py +106 -26
  59. sglang/srt/layers/moe/moe_runner/deep_gemm.py +53 -33
  60. sglang/srt/layers/moe/token_dispatcher/deepep.py +12 -9
  61. sglang/srt/layers/moe/topk.py +31 -6
  62. sglang/srt/layers/pooler.py +21 -2
  63. sglang/srt/layers/quantization/__init__.py +9 -78
  64. sglang/srt/layers/quantization/auto_round.py +394 -0
  65. sglang/srt/layers/quantization/fp8_kernel.py +1 -1
  66. sglang/srt/layers/quantization/fp8_utils.py +2 -2
  67. sglang/srt/layers/quantization/modelopt_quant.py +168 -11
  68. sglang/srt/layers/rotary_embedding.py +117 -45
  69. sglang/srt/lora/lora_registry.py +9 -0
  70. sglang/srt/managers/async_mm_data_processor.py +122 -0
  71. sglang/srt/managers/data_parallel_controller.py +30 -3
  72. sglang/srt/managers/detokenizer_manager.py +3 -0
  73. sglang/srt/managers/io_struct.py +26 -4
  74. sglang/srt/managers/multi_tokenizer_mixin.py +5 -0
  75. sglang/srt/managers/schedule_batch.py +74 -15
  76. sglang/srt/managers/scheduler.py +164 -129
  77. sglang/srt/managers/scheduler_output_processor_mixin.py +40 -3
  78. sglang/srt/managers/scheduler_pp_mixin.py +7 -2
  79. sglang/srt/managers/scheduler_runtime_checker_mixin.py +45 -0
  80. sglang/srt/managers/scheduler_update_weights_mixin.py +18 -3
  81. sglang/srt/managers/session_controller.py +6 -5
  82. sglang/srt/managers/tokenizer_manager.py +154 -59
  83. sglang/srt/managers/tp_worker.py +24 -1
  84. sglang/srt/mem_cache/base_prefix_cache.py +23 -4
  85. sglang/srt/mem_cache/common.py +1 -0
  86. sglang/srt/mem_cache/memory_pool.py +171 -57
  87. sglang/srt/mem_cache/memory_pool_host.py +12 -5
  88. sglang/srt/mem_cache/radix_cache.py +4 -0
  89. sglang/srt/mem_cache/storage/mooncake_store/mooncake_store.py +1 -1
  90. sglang/srt/metrics/collector.py +46 -3
  91. sglang/srt/model_executor/cuda_graph_runner.py +15 -3
  92. sglang/srt/model_executor/forward_batch_info.py +11 -11
  93. sglang/srt/model_executor/model_runner.py +76 -21
  94. sglang/srt/model_executor/npu_graph_runner.py +7 -3
  95. sglang/srt/model_loader/weight_utils.py +1 -1
  96. sglang/srt/models/bailing_moe.py +9 -2
  97. sglang/srt/models/deepseek_nextn.py +11 -2
  98. sglang/srt/models/deepseek_v2.py +149 -34
  99. sglang/srt/models/glm4.py +391 -77
  100. sglang/srt/models/glm4v.py +196 -55
  101. sglang/srt/models/glm4v_moe.py +0 -1
  102. sglang/srt/models/gpt_oss.py +1 -10
  103. sglang/srt/models/kimi_linear.py +678 -0
  104. sglang/srt/models/llama4.py +1 -1
  105. sglang/srt/models/llama_eagle3.py +11 -1
  106. sglang/srt/models/longcat_flash.py +2 -2
  107. sglang/srt/models/minimax_m2.py +1 -1
  108. sglang/srt/models/qwen2.py +1 -1
  109. sglang/srt/models/qwen2_moe.py +30 -15
  110. sglang/srt/models/qwen3.py +1 -1
  111. sglang/srt/models/qwen3_moe.py +16 -8
  112. sglang/srt/models/qwen3_next.py +7 -0
  113. sglang/srt/multimodal/customized_mm_processor_utils.py +35 -0
  114. sglang/srt/multiplex/multiplexing_mixin.py +209 -0
  115. sglang/srt/multiplex/pdmux_context.py +164 -0
  116. sglang/srt/parser/conversation.py +7 -1
  117. sglang/srt/sampling/custom_logit_processor.py +67 -1
  118. sglang/srt/sampling/penaltylib/frequency_penalty.py +6 -8
  119. sglang/srt/sampling/penaltylib/min_new_tokens.py +7 -8
  120. sglang/srt/sampling/penaltylib/orchestrator.py +43 -3
  121. sglang/srt/sampling/penaltylib/presence_penalty.py +6 -8
  122. sglang/srt/server_args.py +103 -22
  123. sglang/srt/single_batch_overlap.py +4 -1
  124. sglang/srt/speculative/draft_utils.py +16 -0
  125. sglang/srt/speculative/eagle_info.py +42 -36
  126. sglang/srt/speculative/eagle_info_v2.py +68 -25
  127. sglang/srt/speculative/eagle_utils.py +261 -16
  128. sglang/srt/speculative/eagle_worker.py +11 -3
  129. sglang/srt/speculative/eagle_worker_v2.py +15 -9
  130. sglang/srt/speculative/spec_info.py +305 -31
  131. sglang/srt/speculative/spec_utils.py +44 -8
  132. sglang/srt/tracing/trace.py +121 -12
  133. sglang/srt/utils/common.py +55 -32
  134. sglang/srt/utils/hf_transformers_utils.py +38 -16
  135. sglang/srt/utils/torch_memory_saver_adapter.py +20 -0
  136. sglang/test/kits/radix_cache_server_kit.py +50 -0
  137. sglang/test/runners.py +31 -7
  138. sglang/test/simple_eval_common.py +5 -3
  139. sglang/test/simple_eval_humaneval.py +1 -0
  140. sglang/test/simple_eval_math.py +1 -0
  141. sglang/test/simple_eval_mmlu.py +1 -0
  142. sglang/test/simple_eval_mmmu_vlm.py +1 -0
  143. sglang/test/test_utils.py +7 -1
  144. sglang/version.py +1 -1
  145. {sglang-0.5.4.post1.dist-info → sglang-0.5.4.post2.dist-info}/METADATA +10 -24
  146. {sglang-0.5.4.post1.dist-info → sglang-0.5.4.post2.dist-info}/RECORD +150 -136
  147. /sglang/test/{kit_matched_stop.py → kits/matched_stop_kit.py} +0 -0
  148. {sglang-0.5.4.post1.dist-info → sglang-0.5.4.post2.dist-info}/WHEEL +0 -0
  149. {sglang-0.5.4.post1.dist-info → sglang-0.5.4.post2.dist-info}/licenses/LICENSE +0 -0
  150. {sglang-0.5.4.post1.dist-info → sglang-0.5.4.post2.dist-info}/top_level.txt +0 -0