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.
- sglang/__init__.py +2 -0
- sglang/api.py +7 -0
- sglang/bench_one_batch.py +8 -6
- sglang/bench_serving.py +1 -1
- sglang/lang/interpreter.py +40 -1
- sglang/lang/ir.py +27 -0
- sglang/math_utils.py +8 -0
- sglang/srt/_custom_ops.py +2 -2
- sglang/srt/code_completion_parser.py +2 -44
- sglang/srt/configs/model_config.py +6 -0
- sglang/srt/constants.py +3 -0
- sglang/srt/conversation.py +19 -3
- sglang/srt/custom_op.py +5 -1
- sglang/srt/disaggregation/base/__init__.py +1 -1
- sglang/srt/disaggregation/base/conn.py +25 -11
- sglang/srt/disaggregation/common/__init__.py +5 -1
- sglang/srt/disaggregation/common/utils.py +42 -0
- sglang/srt/disaggregation/decode.py +211 -72
- sglang/srt/disaggregation/decode_schedule_batch_mixin.py +4 -3
- sglang/srt/disaggregation/fake/__init__.py +1 -1
- sglang/srt/disaggregation/fake/conn.py +15 -9
- sglang/srt/disaggregation/mini_lb.py +34 -4
- sglang/srt/disaggregation/mooncake/__init__.py +1 -1
- sglang/srt/disaggregation/mooncake/conn.py +30 -29
- sglang/srt/disaggregation/nixl/__init__.py +6 -1
- sglang/srt/disaggregation/nixl/conn.py +17 -12
- sglang/srt/disaggregation/prefill.py +144 -55
- sglang/srt/disaggregation/utils.py +155 -123
- sglang/srt/distributed/parallel_state.py +12 -4
- sglang/srt/entrypoints/engine.py +37 -29
- sglang/srt/entrypoints/http_server.py +153 -72
- sglang/srt/entrypoints/http_server_engine.py +0 -3
- sglang/srt/entrypoints/openai/__init__.py +0 -0
- sglang/srt/{openai_api → entrypoints/openai}/protocol.py +84 -10
- sglang/srt/entrypoints/openai/serving_base.py +149 -0
- sglang/srt/entrypoints/openai/serving_chat.py +921 -0
- sglang/srt/entrypoints/openai/serving_completions.py +424 -0
- sglang/srt/entrypoints/openai/serving_embedding.py +169 -0
- sglang/srt/entrypoints/openai/serving_rerank.py +102 -0
- sglang/srt/entrypoints/openai/serving_score.py +61 -0
- sglang/srt/entrypoints/openai/usage_processor.py +81 -0
- sglang/srt/entrypoints/openai/utils.py +72 -0
- sglang/srt/eplb_simulator/__init__.py +1 -0
- sglang/srt/eplb_simulator/reader.py +51 -0
- sglang/srt/function_call/base_format_detector.py +7 -4
- sglang/srt/function_call/deepseekv3_detector.py +1 -1
- sglang/srt/function_call/ebnf_composer.py +64 -10
- sglang/srt/function_call/function_call_parser.py +6 -6
- sglang/srt/function_call/llama32_detector.py +1 -1
- sglang/srt/function_call/mistral_detector.py +1 -1
- sglang/srt/function_call/pythonic_detector.py +1 -1
- sglang/srt/function_call/qwen25_detector.py +1 -1
- sglang/srt/{openai_api/utils.py → jinja_template_utils.py} +6 -5
- sglang/srt/layers/activation.py +40 -3
- sglang/srt/layers/attention/aiter_backend.py +20 -4
- sglang/srt/layers/attention/base_attn_backend.py +1 -1
- sglang/srt/layers/attention/cutlass_mla_backend.py +39 -15
- sglang/srt/layers/attention/flashattention_backend.py +71 -72
- sglang/srt/layers/attention/flashinfer_backend.py +10 -8
- sglang/srt/layers/attention/flashinfer_mla_backend.py +29 -28
- sglang/srt/layers/attention/flashmla_backend.py +7 -12
- sglang/srt/layers/attention/tbo_backend.py +3 -3
- sglang/srt/layers/attention/triton_backend.py +138 -130
- sglang/srt/layers/attention/triton_ops/decode_attention.py +2 -7
- sglang/srt/layers/attention/vision.py +51 -24
- sglang/srt/layers/communicator.py +28 -10
- sglang/srt/layers/dp_attention.py +11 -2
- sglang/srt/layers/layernorm.py +29 -2
- sglang/srt/layers/linear.py +0 -4
- sglang/srt/layers/logits_processor.py +2 -14
- sglang/srt/layers/moe/ep_moe/kernels.py +165 -7
- sglang/srt/layers/moe/ep_moe/layer.py +249 -33
- sglang/srt/layers/moe/ep_moe/token_dispatcher.py +11 -37
- 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
- sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +7 -4
- sglang/srt/layers/moe/fused_moe_triton/layer.py +75 -12
- sglang/srt/layers/moe/topk.py +107 -12
- sglang/srt/layers/pooler.py +56 -0
- sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py +6 -2
- sglang/srt/layers/quantization/deep_gemm_wrapper/__init__.py +1 -0
- sglang/srt/layers/quantization/{deep_gemm.py → deep_gemm_wrapper/compile_utils.py} +23 -80
- sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py +32 -0
- sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py +110 -0
- sglang/srt/layers/quantization/fp8.py +25 -17
- sglang/srt/layers/quantization/fp8_kernel.py +44 -15
- sglang/srt/layers/quantization/fp8_utils.py +87 -22
- sglang/srt/layers/quantization/modelopt_quant.py +62 -8
- sglang/srt/layers/quantization/utils.py +5 -2
- sglang/srt/layers/radix_attention.py +2 -3
- sglang/srt/layers/rotary_embedding.py +42 -2
- sglang/srt/layers/sampler.py +1 -1
- sglang/srt/lora/lora_manager.py +249 -105
- sglang/srt/lora/mem_pool.py +53 -50
- sglang/srt/lora/utils.py +1 -1
- sglang/srt/managers/cache_controller.py +33 -14
- sglang/srt/managers/io_struct.py +31 -10
- sglang/srt/managers/multimodal_processors/base_processor.py +2 -2
- sglang/srt/managers/multimodal_processors/vila.py +85 -0
- sglang/srt/managers/schedule_batch.py +79 -37
- sglang/srt/managers/schedule_policy.py +70 -56
- sglang/srt/managers/scheduler.py +220 -79
- sglang/srt/managers/template_manager.py +226 -0
- sglang/srt/managers/tokenizer_manager.py +40 -10
- sglang/srt/managers/tp_worker.py +12 -2
- sglang/srt/managers/tp_worker_overlap_thread.py +11 -0
- sglang/srt/mem_cache/{paged_allocator.py → allocator.py} +125 -34
- sglang/srt/mem_cache/base_prefix_cache.py +52 -8
- sglang/srt/mem_cache/chunk_cache.py +11 -15
- sglang/srt/mem_cache/hiradix_cache.py +38 -25
- sglang/srt/mem_cache/memory_pool.py +213 -505
- sglang/srt/mem_cache/memory_pool_host.py +380 -0
- sglang/srt/mem_cache/radix_cache.py +56 -28
- sglang/srt/model_executor/cuda_graph_runner.py +198 -100
- sglang/srt/model_executor/forward_batch_info.py +32 -10
- sglang/srt/model_executor/model_runner.py +28 -12
- sglang/srt/model_loader/loader.py +16 -2
- sglang/srt/model_loader/weight_utils.py +11 -2
- sglang/srt/models/bert.py +113 -13
- sglang/srt/models/deepseek_nextn.py +29 -27
- sglang/srt/models/deepseek_v2.py +213 -173
- sglang/srt/models/glm4.py +312 -0
- sglang/srt/models/internvl.py +46 -102
- sglang/srt/models/mimo_mtp.py +2 -18
- sglang/srt/models/roberta.py +117 -9
- sglang/srt/models/vila.py +305 -0
- sglang/srt/reasoning_parser.py +21 -11
- sglang/srt/sampling/sampling_batch_info.py +24 -0
- sglang/srt/sampling/sampling_params.py +2 -0
- sglang/srt/server_args.py +351 -238
- sglang/srt/speculative/build_eagle_tree.py +1 -1
- sglang/srt/speculative/eagle_draft_cuda_graph_runner.py +131 -9
- sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py +130 -14
- sglang/srt/speculative/eagle_utils.py +468 -116
- sglang/srt/speculative/eagle_worker.py +258 -84
- sglang/srt/torch_memory_saver_adapter.py +19 -15
- sglang/srt/two_batch_overlap.py +4 -2
- sglang/srt/utils.py +235 -11
- sglang/test/attention/test_prefix_chunk_info.py +2 -0
- sglang/test/runners.py +38 -3
- sglang/test/test_block_fp8.py +1 -0
- sglang/test/test_block_fp8_deep_gemm_blackwell.py +252 -0
- sglang/test/test_block_fp8_ep.py +2 -0
- sglang/test/test_utils.py +4 -1
- sglang/utils.py +9 -0
- sglang/version.py +1 -1
- {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/METADATA +8 -14
- {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/RECORD +150 -128
- sglang/srt/entrypoints/verl_engine.py +0 -179
- sglang/srt/openai_api/adapter.py +0 -1990
- {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/WHEEL +0 -0
- {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/licenses/LICENSE +0 -0
- {sglang-0.4.7.dist-info → sglang-0.4.8.dist-info}/top_level.txt +0 -0
File without changes
|
File without changes
|
File without changes
|