sglang 0.4.5.post1__py3-none-any.whl → 0.4.5.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 (95) hide show
  1. sglang/__init__.py +2 -4
  2. sglang/bench_one_batch.py +2 -2
  3. sglang/bench_serving.py +0 -4
  4. sglang/lang/backend/anthropic.py +0 -4
  5. sglang/lang/backend/base_backend.py +1 -1
  6. sglang/lang/backend/openai.py +1 -1
  7. sglang/lang/backend/vertexai.py +0 -1
  8. sglang/lang/compiler.py +1 -7
  9. sglang/lang/tracer.py +3 -7
  10. sglang/srt/_custom_ops.py +0 -2
  11. sglang/srt/constrained/outlines_jump_forward.py +14 -1
  12. sglang/srt/constrained/triton_ops/bitmask_ops.py +141 -0
  13. sglang/srt/constrained/xgrammar_backend.py +26 -4
  14. sglang/srt/custom_op.py +0 -62
  15. sglang/srt/disaggregation/decode.py +62 -6
  16. sglang/srt/disaggregation/mini_lb.py +5 -1
  17. sglang/srt/disaggregation/mooncake/conn.py +32 -62
  18. sglang/srt/disaggregation/mooncake/transfer_engine.py +30 -61
  19. sglang/srt/disaggregation/prefill.py +40 -4
  20. sglang/srt/disaggregation/utils.py +15 -0
  21. sglang/srt/entrypoints/verl_engine.py +7 -5
  22. sglang/srt/layers/activation.py +6 -8
  23. sglang/srt/layers/attention/flashattention_backend.py +114 -71
  24. sglang/srt/layers/attention/flashinfer_backend.py +5 -2
  25. sglang/srt/layers/attention/torch_native_backend.py +6 -1
  26. sglang/srt/layers/attention/triton_backend.py +6 -0
  27. sglang/srt/layers/attention/triton_ops/extend_attention.py +13 -2
  28. sglang/srt/layers/layernorm.py +1 -1
  29. sglang/srt/layers/linear.py +17 -3
  30. sglang/srt/layers/moe/ep_moe/layer.py +15 -29
  31. sglang/srt/layers/moe/fused_moe_native.py +4 -0
  32. sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +14 -19
  33. sglang/srt/layers/moe/fused_moe_triton/layer.py +7 -0
  34. sglang/srt/layers/moe/topk.py +27 -30
  35. sglang/srt/layers/parameter.py +0 -2
  36. sglang/srt/layers/quantization/__init__.py +1 -0
  37. sglang/srt/layers/quantization/blockwise_int8.py +2 -0
  38. sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py +8 -2
  39. sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py +16 -44
  40. sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py +4 -7
  41. sglang/srt/layers/quantization/fp8.py +115 -132
  42. sglang/srt/layers/quantization/fp8_kernel.py +213 -57
  43. sglang/srt/layers/quantization/fp8_utils.py +187 -262
  44. sglang/srt/layers/quantization/moe_wna16.py +2 -0
  45. sglang/srt/layers/quantization/utils.py +5 -11
  46. sglang/srt/layers/quantization/w8a8_fp8.py +2 -0
  47. sglang/srt/layers/quantization/w8a8_int8.py +7 -7
  48. sglang/srt/layers/radix_attention.py +15 -0
  49. sglang/srt/layers/rotary_embedding.py +3 -2
  50. sglang/srt/layers/sampler.py +5 -10
  51. sglang/srt/lora/backend/base_backend.py +18 -2
  52. sglang/srt/lora/backend/flashinfer_backend.py +1 -1
  53. sglang/srt/lora/backend/triton_backend.py +1 -1
  54. sglang/srt/lora/layers.py +1 -1
  55. sglang/srt/lora/lora.py +1 -1
  56. sglang/srt/lora/lora_manager.py +1 -1
  57. sglang/srt/managers/detokenizer_manager.py +0 -1
  58. sglang/srt/managers/io_struct.py +1 -0
  59. sglang/srt/managers/mm_utils.py +4 -3
  60. sglang/srt/managers/multimodal_processor.py +0 -2
  61. sglang/srt/managers/multimodal_processors/base_processor.py +3 -2
  62. sglang/srt/managers/schedule_batch.py +2 -4
  63. sglang/srt/managers/scheduler.py +12 -71
  64. sglang/srt/managers/tokenizer_manager.py +1 -0
  65. sglang/srt/mem_cache/hiradix_cache.py +5 -1
  66. sglang/srt/mem_cache/memory_pool.py +7 -2
  67. sglang/srt/model_executor/cuda_graph_runner.py +2 -2
  68. sglang/srt/model_executor/model_runner.py +20 -27
  69. sglang/srt/models/bert.py +398 -0
  70. sglang/srt/models/deepseek.py +1 -1
  71. sglang/srt/models/deepseek_nextn.py +74 -70
  72. sglang/srt/models/deepseek_v2.py +289 -348
  73. sglang/srt/models/llama.py +5 -5
  74. sglang/srt/models/minicpm3.py +29 -201
  75. sglang/srt/models/qwen2.py +4 -1
  76. sglang/srt/models/qwen2_moe.py +14 -13
  77. sglang/srt/models/qwen3.py +335 -0
  78. sglang/srt/models/qwen3_moe.py +423 -0
  79. sglang/srt/reasoning_parser.py +0 -1
  80. sglang/srt/sampling/sampling_batch_info.py +2 -3
  81. sglang/srt/server_args.py +34 -32
  82. sglang/srt/speculative/eagle_worker.py +4 -7
  83. sglang/srt/utils.py +16 -1
  84. sglang/test/runners.py +5 -1
  85. sglang/test/test_block_fp8.py +167 -0
  86. sglang/test/test_custom_ops.py +1 -1
  87. sglang/version.py +1 -1
  88. {sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/METADATA +3 -3
  89. {sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/RECORD +92 -91
  90. {sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/WHEEL +1 -1
  91. sglang/lang/__init__.py +0 -0
  92. sglang/srt/lora/backend/__init__.py +0 -25
  93. sglang/srt/server.py +0 -18
  94. {sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/licenses/LICENSE +0 -0
  95. {sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/top_level.txt +0 -0
sglang/srt/lora/backend/__init__.py DELETED
@@ -1,25 +0,0 @@
1
- from sglang.srt.lora.backend.base_backend import BaseLoRABackend
2
-
3
-
4
- def get_backend_from_name(name: str) -> BaseLoRABackend:
5
- """
6
- Get corresponding backend class from backend's name
7
- """
8
- if name == "triton":
9
- from sglang.srt.lora.backend.triton_backend import TritonLoRABackend
10
-
11
- return TritonLoRABackend
12
- elif name == "flashinfer":
13
- from sglang.srt.lora.backend.flashinfer_backend import FlashInferLoRABackend
14
-
15
- return FlashInferLoRABackend
16
- else:
17
- raise ValueError(f"Invalid backend: {name}")
18
-
19
-
20
- __all__ = [
21
- "BaseLoRABackend",
22
- "FlashInferLoRABackend",
23
- "TritonLoRABackend",
24
- "get_backend_from_name",
25
- ]
sglang/srt/server.py DELETED
@@ -1,18 +0,0 @@
1
- # Copyright 2023-2024 SGLang Team
2
- # Licensed under the Apache License, Version 2.0 (the "License");
3
- # you may not use this file except in compliance with the License.
4
- # You may obtain a copy of the License at
5
- #
6
- # http://www.apache.org/licenses/LICENSE-2.0
7
- #
8
- # Unless required by applicable law or agreed to in writing, software
9
- # distributed under the License is distributed on an "AS IS" BASIS,
10
- # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
11
- # See the License for the specific language governing permissions and
12
- # limitations under the License.
13
- # ==============================================================================
14
-
15
- # Some shortcuts for backward compatibility.
16
- # They will be removed in new versions.
17
- from sglang.srt.entrypoints.engine import Engine
18
- from sglang.srt.entrypoints.http_server import kill_process_tree, launch_server