sglang 0.5.2rc2__py3-none-any.whl → 0.5.3rc2__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.
- sglang/bench_one_batch.py +7 -9
- sglang/bench_one_batch_server.py +330 -31
- sglang/bench_serving.py +267 -32
- sglang/global_config.py +2 -2
- sglang/lang/backend/runtime_endpoint.py +1 -1
- sglang/launch_server.py +14 -0
- sglang/profiler.py +2 -2
- sglang/srt/batch_invariant_ops/__init__.py +27 -0
- sglang/srt/batch_invariant_ops/batch_invariant_ops.py +549 -0
- sglang/srt/configs/__init__.py +8 -0
- sglang/srt/configs/device_config.py +3 -1
- sglang/srt/configs/dots_ocr.py +64 -0
- sglang/srt/configs/dots_vlm.py +139 -0
- sglang/srt/configs/falcon_h1.py +360 -0
- sglang/srt/configs/load_config.py +9 -0
- sglang/srt/configs/model_config.py +181 -82
- sglang/srt/configs/qwen3_next.py +326 -0
- sglang/srt/configs/qwen3_vl.py +586 -0
- sglang/srt/connector/__init__.py +8 -1
- sglang/srt/connector/remote_instance.py +82 -0
- sglang/srt/constrained/base_grammar_backend.py +49 -12
- sglang/srt/constrained/llguidance_backend.py +0 -1
- sglang/srt/constrained/outlines_backend.py +0 -1
- sglang/srt/constrained/outlines_jump_forward.py +1 -1
- sglang/srt/constrained/xgrammar_backend.py +30 -9
- sglang/srt/custom_op.py +11 -1
- sglang/srt/debug_utils/dump_comparator.py +81 -44
- sglang/srt/debug_utils/dump_loader.py +97 -0
- sglang/srt/debug_utils/dumper.py +21 -6
- sglang/srt/debug_utils/text_comparator.py +73 -11
- sglang/srt/disaggregation/ascend/conn.py +2 -2
- sglang/srt/disaggregation/ascend/transfer_engine.py +47 -9
- sglang/srt/disaggregation/base/conn.py +1 -1
- sglang/srt/disaggregation/common/conn.py +279 -108
- sglang/srt/disaggregation/decode.py +71 -19
- sglang/srt/disaggregation/decode_kvcache_offload_manager.py +185 -0
- sglang/srt/disaggregation/decode_schedule_batch_mixin.py +29 -17
- sglang/srt/disaggregation/fake/conn.py +1 -1
- sglang/srt/disaggregation/mini_lb.py +6 -445
- sglang/srt/disaggregation/mooncake/conn.py +55 -537
- sglang/srt/disaggregation/nixl/conn.py +326 -53
- sglang/srt/disaggregation/prefill.py +36 -17
- sglang/srt/disaggregation/utils.py +40 -54
- sglang/srt/distributed/device_communicators/all_reduce_utils.py +16 -0
- sglang/srt/distributed/device_communicators/shm_broadcast.py +4 -2
- sglang/srt/distributed/device_communicators/symm_mem.py +164 -0
- sglang/srt/distributed/parallel_state.py +156 -80
- sglang/srt/entrypoints/engine.py +59 -18
- sglang/srt/entrypoints/grpc_request_manager.py +855 -0
- sglang/srt/entrypoints/grpc_server.py +810 -0
- sglang/srt/entrypoints/http_server.py +130 -59
- sglang/srt/entrypoints/openai/protocol.py +112 -4
- sglang/srt/entrypoints/openai/serving_base.py +65 -3
- sglang/srt/entrypoints/openai/serving_chat.py +204 -55
- sglang/srt/entrypoints/openai/serving_completions.py +14 -3
- sglang/srt/entrypoints/openai/serving_embedding.py +9 -3
- sglang/srt/entrypoints/openai/serving_rerank.py +3 -1
- sglang/srt/entrypoints/openai/serving_responses.py +48 -3
- sglang/srt/entrypoints/openai/serving_score.py +1 -0
- sglang/srt/environ.py +285 -0
- sglang/srt/eplb/eplb_manager.py +2 -2
- sglang/srt/eplb/expert_distribution.py +26 -13
- sglang/srt/eplb/expert_location.py +38 -8
- sglang/srt/eplb/expert_location_updater.py +1 -1
- sglang/srt/function_call/base_format_detector.py +3 -6
- sglang/srt/function_call/ebnf_composer.py +11 -9
- sglang/srt/function_call/function_call_parser.py +9 -2
- sglang/srt/function_call/glm4_moe_detector.py +4 -4
- sglang/srt/function_call/gpt_oss_detector.py +23 -0
- sglang/srt/function_call/json_array_parser.py +63 -0
- sglang/srt/function_call/kimik2_detector.py +17 -4
- sglang/srt/function_call/qwen3_coder_detector.py +1 -1
- sglang/srt/function_call/utils.py +96 -5
- sglang/srt/grpc/__init__.py +1 -0
- sglang/srt/grpc/compile_proto.py +245 -0
- sglang/srt/grpc/sglang_scheduler_pb2.py +111 -0
- sglang/srt/grpc/sglang_scheduler_pb2.pyi +434 -0
- sglang/srt/grpc/sglang_scheduler_pb2_grpc.py +239 -0
- sglang/srt/layers/activation.py +143 -9
- sglang/srt/layers/attention/aiter_backend.py +14 -15
- sglang/srt/layers/attention/ascend_backend.py +115 -9
- sglang/srt/layers/attention/attention_registry.py +206 -0
- sglang/srt/layers/attention/base_attn_backend.py +12 -3
- sglang/srt/layers/attention/cutlass_mla_backend.py +3 -3
- sglang/srt/layers/attention/dual_chunk_flashattention_backend.py +1 -1
- sglang/srt/layers/attention/fla/chunk.py +242 -0
- sglang/srt/layers/attention/fla/chunk_delta_h.py +314 -0
- sglang/srt/layers/attention/fla/chunk_o.py +178 -0
- sglang/srt/layers/attention/fla/chunk_scaled_dot_kkt.py +151 -0
- sglang/srt/layers/attention/fla/cumsum.py +300 -0
- sglang/srt/layers/attention/fla/fused_recurrent.py +640 -0
- sglang/srt/layers/attention/fla/fused_sigmoid_gating_recurrent.py +232 -0
- sglang/srt/layers/attention/fla/index.py +37 -0
- sglang/srt/layers/attention/fla/l2norm.py +150 -0
- sglang/srt/layers/attention/fla/layernorm_gated.py +326 -0
- sglang/srt/layers/attention/fla/op.py +66 -0
- sglang/srt/layers/attention/fla/solve_tril.py +465 -0
- sglang/srt/layers/attention/fla/utils.py +331 -0
- sglang/srt/layers/attention/fla/wy_fast.py +158 -0
- sglang/srt/layers/attention/flashattention_backend.py +41 -8
- sglang/srt/layers/attention/flashinfer_backend.py +118 -198
- sglang/srt/layers/attention/flashinfer_mla_backend.py +27 -27
- sglang/srt/layers/attention/flashmla_backend.py +7 -5
- sglang/srt/layers/attention/hybrid_attn_backend.py +68 -53
- sglang/srt/layers/attention/hybrid_linear_attn_backend.py +602 -0
- sglang/srt/layers/attention/intel_amx_backend.py +3 -0
- sglang/srt/layers/attention/mamba/causal_conv1d.py +129 -0
- sglang/srt/layers/attention/mamba/causal_conv1d_triton.py +969 -0
- sglang/srt/layers/attention/mamba/mamba.py +629 -0
- sglang/srt/layers/attention/mamba/mamba_utils.py +81 -0
- sglang/srt/layers/attention/mamba/ops/__init__.py +2 -0
- sglang/srt/layers/attention/mamba/ops/layernorm_gated.py +172 -0
- sglang/srt/layers/attention/mamba/ops/mamba_ssm.py +442 -0
- sglang/srt/layers/attention/mamba/ops/ssd_bmm.py +264 -0
- sglang/srt/layers/attention/mamba/ops/ssd_chunk_scan.py +622 -0
- sglang/srt/layers/attention/mamba/ops/ssd_chunk_state.py +757 -0
- sglang/srt/layers/attention/mamba/ops/ssd_combined.py +262 -0
- sglang/srt/layers/attention/mamba/ops/ssd_state_passing.py +275 -0
- sglang/srt/layers/attention/npu_ops/mla_preprocess.py +393 -0
- sglang/srt/layers/attention/nsa/dequant_k_cache.py +163 -0
- sglang/srt/layers/attention/nsa/index_buf_accessor.py +354 -0
- sglang/srt/layers/attention/nsa/nsa_indexer.py +761 -0
- sglang/srt/layers/attention/nsa/quant_k_cache.py +255 -0
- sglang/srt/layers/attention/nsa/tilelang_kernel.py +785 -0
- sglang/srt/layers/attention/nsa/transform_index.py +144 -0
- sglang/srt/layers/attention/nsa/utils.py +24 -0
- sglang/srt/layers/attention/nsa_backend.py +887 -0
- sglang/srt/layers/attention/tbo_backend.py +6 -6
- sglang/srt/layers/attention/torch_flex_backend.py +325 -0
- sglang/srt/layers/attention/torch_native_backend.py +12 -6
- sglang/srt/layers/attention/triton_backend.py +57 -7
- sglang/srt/layers/attention/trtllm_mha_backend.py +5 -7
- sglang/srt/layers/attention/trtllm_mla_backend.py +276 -39
- sglang/srt/layers/attention/vision.py +58 -0
- sglang/srt/layers/attention/wave_backend.py +4 -4
- sglang/srt/layers/attention/wave_ops/decode_attention.py +2 -4
- sglang/srt/layers/attention/wave_ops/extend_attention.py +1 -3
- sglang/srt/layers/communicator.py +8 -0
- sglang/srt/layers/dp_attention.py +41 -2
- sglang/srt/layers/elementwise.py +3 -1
- sglang/srt/layers/layernorm.py +34 -15
- sglang/srt/layers/linear.py +55 -7
- sglang/srt/layers/logits_processor.py +44 -12
- sglang/srt/layers/moe/__init__.py +2 -1
- sglang/srt/layers/moe/cutlass_w4a8_moe.py +3 -3
- sglang/srt/layers/moe/ep_moe/kernels.py +2 -2
- sglang/srt/layers/moe/ep_moe/layer.py +256 -63
- sglang/srt/layers/moe/flashinfer_cutedsl_moe.py +183 -0
- sglang/srt/layers/moe/fused_moe_native.py +5 -3
- 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
- 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
- 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
- sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=512,device_name=NVIDIA_H20.json +146 -0
- 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
- sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H20-3e.json +146 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=128,device_name=NVIDIA_H200.json +146 -0
- 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
- sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_B200.json +146 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H20-3e.json +146 -0
- sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=256,device_name=NVIDIA_H200.json +146 -0
- 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
- sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=512,N=64,device_name=NVIDIA_H200.json +146 -0
- sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +5 -2
- sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_config.py +7 -3
- sglang/srt/layers/moe/fused_moe_triton/fused_moe_triton_kernels.py +23 -20
- sglang/srt/layers/moe/fused_moe_triton/layer.py +71 -70
- sglang/srt/layers/moe/moe_runner/__init__.py +2 -1
- sglang/srt/layers/moe/moe_runner/base.py +274 -1
- sglang/srt/layers/moe/moe_runner/runner.py +80 -0
- sglang/srt/layers/moe/moe_runner/triton.py +448 -0
- sglang/srt/layers/moe/token_dispatcher/__init__.py +16 -4
- sglang/srt/layers/moe/token_dispatcher/{base_dispatcher.py → base.py} +67 -17
- sglang/srt/layers/moe/token_dispatcher/deepep.py +118 -56
- sglang/srt/layers/moe/token_dispatcher/standard.py +44 -2
- sglang/srt/layers/moe/topk.py +30 -9
- sglang/srt/layers/moe/utils.py +22 -6
- sglang/srt/layers/parameter.py +23 -6
- sglang/srt/layers/quantization/awq.py +19 -7
- sglang/srt/layers/quantization/base_config.py +11 -6
- sglang/srt/layers/quantization/blockwise_int8.py +38 -27
- sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py +1 -0
- sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py +50 -30
- sglang/srt/layers/quantization/compressed_tensors/schemes/__init__.py +2 -0
- sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py +13 -1
- sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_int8.py +173 -0
- sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py +2 -10
- sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py +27 -0
- sglang/srt/layers/quantization/fp8.py +78 -49
- sglang/srt/layers/quantization/fp8_utils.py +51 -32
- sglang/srt/layers/quantization/gptq.py +25 -17
- sglang/srt/layers/quantization/modelopt_quant.py +190 -55
- sglang/srt/layers/quantization/moe_wna16.py +21 -18
- sglang/srt/layers/quantization/mxfp4.py +74 -42
- sglang/srt/layers/quantization/quark/quark_moe.py +48 -30
- sglang/srt/layers/quantization/unquant.py +135 -47
- sglang/srt/layers/quantization/w4afp8.py +26 -17
- sglang/srt/layers/quantization/w8a8_fp8.py +35 -20
- sglang/srt/layers/quantization/w8a8_int8.py +91 -41
- sglang/srt/layers/rotary_embedding.py +78 -31
- sglang/srt/layers/sampler.py +213 -21
- sglang/srt/layers/utils.py +23 -0
- sglang/srt/lora/backend/base_backend.py +50 -8
- sglang/srt/lora/backend/chunked_backend.py +348 -0
- sglang/srt/lora/backend/triton_backend.py +99 -5
- sglang/srt/lora/layers.py +32 -0
- sglang/srt/lora/lora.py +8 -3
- sglang/srt/lora/lora_manager.py +52 -118
- sglang/srt/lora/mem_pool.py +25 -11
- sglang/srt/lora/triton_ops/__init__.py +4 -0
- sglang/srt/lora/triton_ops/chunked_sgmv_expand.py +214 -0
- sglang/srt/lora/triton_ops/chunked_sgmv_shrink.py +174 -0
- sglang/srt/lora/utils.py +22 -11
- sglang/srt/managers/async_dynamic_batch_tokenizer.py +170 -0
- sglang/srt/managers/cache_controller.py +199 -301
- sglang/srt/managers/data_parallel_controller.py +115 -80
- sglang/srt/managers/detokenizer_manager.py +19 -15
- sglang/srt/managers/disagg_service.py +46 -0
- sglang/srt/managers/io_struct.py +340 -109
- sglang/srt/managers/mm_utils.py +44 -6
- sglang/srt/managers/multi_tokenizer_mixin.py +357 -407
- sglang/srt/managers/multimodal_processor.py +1 -2
- sglang/srt/managers/overlap_utils.py +53 -0
- sglang/srt/managers/schedule_batch.py +240 -138
- sglang/srt/managers/schedule_policy.py +144 -17
- sglang/srt/managers/scheduler.py +502 -209
- sglang/srt/managers/scheduler_input_blocker.py +1 -1
- sglang/srt/managers/scheduler_metrics_mixin.py +99 -126
- sglang/srt/managers/scheduler_output_processor_mixin.py +75 -22
- sglang/srt/managers/scheduler_profiler_mixin.py +6 -6
- sglang/srt/managers/scheduler_update_weights_mixin.py +7 -0
- sglang/srt/managers/tokenizer_communicator_mixin.py +675 -0
- sglang/srt/managers/tokenizer_manager.py +320 -632
- sglang/srt/managers/tp_worker.py +81 -22
- sglang/srt/managers/tp_worker_overlap_thread.py +71 -56
- sglang/srt/managers/utils.py +1 -45
- sglang/srt/mem_cache/allocator.py +14 -20
- sglang/srt/mem_cache/allocator_ascend.py +41 -27
- sglang/srt/mem_cache/base_prefix_cache.py +1 -1
- sglang/srt/mem_cache/chunk_cache.py +8 -1
- sglang/srt/mem_cache/evict_policy.py +23 -0
- sglang/srt/mem_cache/hicache_storage.py +43 -24
- sglang/srt/mem_cache/hiradix_cache.py +222 -75
- sglang/srt/mem_cache/memory_pool.py +535 -58
- sglang/srt/mem_cache/memory_pool_host.py +239 -228
- sglang/srt/mem_cache/radix_cache.py +222 -73
- sglang/srt/mem_cache/radix_cache_cpp.py +11 -8
- sglang/srt/mem_cache/storage/__init__.py +10 -0
- sglang/srt/mem_cache/storage/aibrix_kvcache/aibrix_kvcache_storage.py +151 -0
- sglang/srt/mem_cache/storage/aibrix_kvcache/unit_test.py +109 -0
- sglang/srt/mem_cache/storage/backend_factory.py +223 -0
- sglang/srt/mem_cache/storage/eic/eic_storage.py +778 -0
- sglang/srt/mem_cache/storage/eic/test_unit.py +115 -0
- sglang/srt/mem_cache/storage/hf3fs/hf3fs_client.py +164 -0
- sglang/srt/mem_cache/storage/hf3fs/{client_hf3fs.py → hf3fs_usrbio_client.py} +5 -1
- sglang/srt/mem_cache/storage/hf3fs/storage_hf3fs.py +259 -62
- sglang/srt/mem_cache/storage/lmcache/lmc_radix_cache.py +284 -0
- sglang/srt/mem_cache/storage/lmcache/unit_test.py +121 -0
- sglang/srt/mem_cache/storage/mooncake_store/mooncake_store.py +166 -17
- sglang/srt/mem_cache/swa_radix_cache.py +25 -36
- sglang/srt/metrics/collector.py +511 -132
- sglang/srt/metrics/func_timer.py +2 -7
- sglang/srt/metrics/startup_func_log_and_timer.py +150 -0
- sglang/srt/metrics/utils.py +8 -1
- sglang/srt/model_executor/cpu_graph_runner.py +640 -0
- sglang/srt/model_executor/cuda_graph_runner.py +52 -37
- sglang/srt/model_executor/forward_batch_info.py +82 -40
- sglang/srt/model_executor/model_runner.py +432 -157
- sglang/srt/model_executor/npu_graph_runner.py +12 -5
- sglang/srt/model_loader/__init__.py +9 -3
- sglang/srt/model_loader/loader.py +133 -5
- sglang/srt/model_loader/remote_instance_weight_loader_utils.py +69 -0
- sglang/srt/model_loader/weight_utils.py +158 -3
- sglang/srt/models/apertus.py +686 -0
- sglang/srt/models/bailing_moe.py +820 -217
- sglang/srt/models/bailing_moe_nextn.py +168 -0
- sglang/srt/models/deepseek_nextn.py +6 -1
- sglang/srt/models/deepseek_v2.py +607 -130
- sglang/srt/models/dots_ocr.py +173 -0
- sglang/srt/models/dots_vlm.py +174 -0
- sglang/srt/models/dots_vlm_vit.py +337 -0
- sglang/srt/models/ernie4.py +1 -1
- sglang/srt/models/falcon_h1.py +576 -0
- sglang/srt/models/gemma3_causal.py +0 -2
- sglang/srt/models/gemma3_mm.py +1 -1
- sglang/srt/models/gemma3n_mm.py +2 -2
- sglang/srt/models/glm4_moe.py +4 -4
- sglang/srt/models/glm4_moe_nextn.py +2 -2
- sglang/srt/models/glm4v.py +5 -3
- sglang/srt/models/glm4v_moe.py +4 -1
- sglang/srt/models/gpt_oss.py +8 -31
- sglang/srt/models/kimi_vl_moonvit.py +2 -2
- sglang/srt/models/llama.py +4 -0
- sglang/srt/models/llama4.py +9 -0
- sglang/srt/models/llama_eagle3.py +13 -0
- sglang/srt/models/longcat_flash.py +3 -3
- sglang/srt/models/longcat_flash_nextn.py +1 -1
- sglang/srt/models/mllama4.py +40 -4
- sglang/srt/models/opt.py +637 -0
- sglang/srt/models/qwen2_5_vl.py +29 -5
- sglang/srt/models/qwen2_audio.py +1 -1
- sglang/srt/models/qwen2_moe.py +120 -13
- sglang/srt/models/qwen2_vl.py +1 -1
- sglang/srt/models/qwen3.py +18 -3
- sglang/srt/models/qwen3_moe.py +32 -4
- sglang/srt/models/qwen3_next.py +1069 -0
- sglang/srt/models/qwen3_next_mtp.py +112 -0
- sglang/srt/models/qwen3_vl.py +787 -0
- sglang/srt/models/qwen3_vl_moe.py +471 -0
- sglang/srt/models/registry.py +15 -3
- sglang/srt/models/sarashina2_vision.py +269 -0
- sglang/srt/models/solar.py +505 -0
- sglang/srt/models/starcoder2.py +357 -0
- sglang/srt/models/step3_vl.py +1 -1
- sglang/srt/models/torch_native_llama.py +9 -2
- sglang/srt/models/utils.py +51 -0
- sglang/srt/multimodal/processors/base_processor.py +15 -7
- sglang/srt/multimodal/processors/dots_vlm.py +98 -0
- sglang/srt/multimodal/processors/glm4v.py +9 -9
- sglang/srt/multimodal/processors/internvl.py +153 -129
- sglang/srt/multimodal/processors/qwen_vl.py +23 -6
- sglang/srt/multimodal/processors/sarashina2_vision.py +81 -0
- sglang/srt/offloader.py +27 -3
- sglang/srt/parser/jinja_template_utils.py +6 -0
- sglang/srt/sampling/sampling_batch_info.py +38 -17
- sglang/srt/sampling/sampling_params.py +7 -0
- sglang/srt/server_args.py +966 -267
- sglang/srt/server_args_config_parser.py +146 -0
- sglang/srt/single_batch_overlap.py +151 -0
- sglang/srt/speculative/cpp_ngram/ngram.cpp +374 -0
- sglang/srt/speculative/cpp_ngram/ngram.h +110 -0
- sglang/srt/speculative/cpp_ngram/ngram_cache.py +138 -0
- sglang/srt/speculative/cpp_ngram/ngram_cache_binding.cpp +43 -0
- sglang/srt/speculative/cpp_ngram/param.h +125 -0
- sglang/srt/speculative/cpp_ngram/queue.h +71 -0
- sglang/srt/speculative/eagle_draft_cuda_graph_runner.py +7 -1
- sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py +13 -2
- sglang/srt/speculative/{eagle_utils.py → eagle_info.py} +207 -757
- sglang/srt/speculative/eagle_worker.py +99 -28
- sglang/srt/speculative/ngram_utils.py +428 -0
- sglang/srt/speculative/ngram_worker.py +245 -0
- sglang/srt/speculative/spec_info.py +52 -0
- sglang/srt/speculative/spec_utils.py +606 -0
- sglang/srt/speculative/standalone_worker.py +109 -0
- sglang/srt/torch_memory_saver_adapter.py +5 -7
- sglang/srt/tracing/trace.py +578 -0
- sglang/srt/two_batch_overlap.py +8 -5
- sglang/srt/utils/__init__.py +2 -0
- sglang/srt/{utils.py → utils/common.py} +433 -77
- sglang/srt/{hf_transformers_utils.py → utils/hf_transformers_utils.py} +53 -5
- sglang/srt/{patch_torch.py → utils/patch_torch.py} +8 -0
- sglang/srt/utils/rpd_utils.py +452 -0
- sglang/srt/utils/slow_rank_detector.py +71 -0
- sglang/srt/warmup.py +8 -4
- sglang/srt/weight_sync/utils.py +2 -2
- sglang/test/attention/test_trtllm_mla_backend.py +169 -5
- sglang/test/get_logits_ut.py +57 -0
- sglang/test/run_eval.py +79 -11
- sglang/test/runners.py +5 -1
- sglang/test/simple_eval_common.py +5 -2
- sglang/test/simple_eval_mmmu_vlm.py +441 -0
- sglang/test/test_block_fp8.py +2 -2
- sglang/test/test_cutlass_moe.py +24 -6
- sglang/test/test_deterministic.py +297 -0
- sglang/test/test_disaggregation_utils.py +77 -0
- sglang/test/test_fp4_moe.py +370 -1
- sglang/test/test_programs.py +1 -1
- sglang/test/test_utils.py +383 -5
- sglang/utils.py +21 -1
- sglang/version.py +1 -1
- {sglang-0.5.2rc2.dist-info → sglang-0.5.3rc2.dist-info}/METADATA +69 -124
- {sglang-0.5.2rc2.dist-info → sglang-0.5.3rc2.dist-info}/RECORD +375 -245
- sglang/srt/disaggregation/launch_lb.py +0 -118
- sglang/srt/mem_cache/lora_radix_cache.py +0 -421
- /sglang/srt/{poll_based_barrier.py → utils/poll_based_barrier.py} +0 -0
- {sglang-0.5.2rc2.dist-info → sglang-0.5.3rc2.dist-info}/WHEEL +0 -0
- {sglang-0.5.2rc2.dist-info → sglang-0.5.3rc2.dist-info}/licenses/LICENSE +0 -0
- {sglang-0.5.2rc2.dist-info → sglang-0.5.3rc2.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.
|
4
|
-
Summary: SGLang is
|
3
|
+
Version: 0.5.3rc2
|
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:
|
216
|
-
Requires-Dist:
|
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:
|
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
|
-
|
221
|
-
Requires-Dist:
|
222
|
-
Requires-Dist:
|
223
|
-
Requires-Dist:
|
224
|
-
Requires-Dist:
|
225
|
-
Requires-Dist:
|
226
|
-
Requires-Dist:
|
227
|
-
Requires-Dist:
|
228
|
-
Requires-Dist:
|
229
|
-
Requires-Dist:
|
230
|
-
Requires-Dist:
|
231
|
-
Requires-Dist:
|
232
|
-
Requires-Dist:
|
233
|
-
Requires-Dist:
|
234
|
-
Requires-Dist:
|
235
|
-
Requires-Dist:
|
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[
|
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-
|
315
|
-
Requires-Dist: sglang[
|
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
|
-
|
343
|
-
|
344
|
-
Requires-Dist: sglang[test]; extra == "
|
345
|
-
|
346
|
-
|
347
|
-
Requires-Dist: sglang[test]; extra == "
|
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">
|