sglang 0.4.9.post6__py3-none-any.whl → 0.4.10__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 (69) hide show
  1. sglang/bench_one_batch.py +3 -0
  2. sglang/srt/configs/__init__.py +8 -0
  3. sglang/srt/configs/model_config.py +3 -0
  4. sglang/srt/configs/step3_vl.py +172 -0
  5. sglang/srt/conversation.py +23 -0
  6. sglang/srt/disaggregation/decode.py +2 -8
  7. sglang/srt/disaggregation/prefill.py +2 -6
  8. sglang/srt/distributed/parallel_state.py +86 -1
  9. sglang/srt/entrypoints/engine.py +14 -18
  10. sglang/srt/entrypoints/http_server.py +10 -2
  11. sglang/srt/entrypoints/openai/serving_chat.py +2 -21
  12. sglang/srt/eplb/expert_distribution.py +5 -0
  13. sglang/srt/eplb/expert_location.py +17 -6
  14. sglang/srt/eplb/expert_location_dispatch.py +1 -0
  15. sglang/srt/eplb/expert_location_updater.py +2 -0
  16. sglang/srt/function_call/function_call_parser.py +2 -0
  17. sglang/srt/function_call/step3_detector.py +436 -0
  18. sglang/srt/hf_transformers_utils.py +2 -0
  19. sglang/srt/jinja_template_utils.py +4 -1
  20. sglang/srt/layers/moe/cutlass_moe.py +2 -1
  21. sglang/srt/layers/moe/ep_moe/layer.py +20 -640
  22. sglang/srt/layers/moe/fused_moe_triton/fused_moe.py +26 -13
  23. sglang/srt/layers/moe/fused_moe_triton/layer.py +97 -38
  24. sglang/srt/layers/quantization/fp8.py +0 -18
  25. sglang/srt/layers/quantization/unquant.py +0 -8
  26. sglang/srt/layers/quantization/w4afp8.py +1 -0
  27. sglang/srt/managers/cache_controller.py +143 -45
  28. sglang/srt/managers/data_parallel_controller.py +2 -0
  29. sglang/srt/managers/io_struct.py +0 -2
  30. sglang/srt/managers/scheduler.py +89 -671
  31. sglang/srt/managers/scheduler_metrics_mixin.py +229 -0
  32. sglang/srt/managers/scheduler_profiler_mixin.py +279 -0
  33. sglang/srt/managers/scheduler_update_weights_mixin.py +142 -0
  34. sglang/srt/managers/template_manager.py +62 -19
  35. sglang/srt/managers/tokenizer_manager.py +123 -74
  36. sglang/srt/managers/tp_worker.py +4 -0
  37. sglang/srt/managers/tp_worker_overlap_thread.py +2 -1
  38. sglang/srt/mem_cache/hicache_storage.py +45 -11
  39. sglang/srt/mem_cache/hiradix_cache.py +15 -4
  40. sglang/srt/mem_cache/memory_pool_host.py +73 -1
  41. sglang/srt/mem_cache/mooncake_store/mooncake_store.py +264 -0
  42. sglang/srt/mem_cache/mooncake_store/unit_test.py +40 -0
  43. sglang/srt/mem_cache/storage/hf3fs/client_hf3fs.py +177 -0
  44. sglang/srt/mem_cache/storage/hf3fs/storage_hf3fs.py +278 -0
  45. sglang/srt/mem_cache/storage/hf3fs/test_hf3fs_utils.py +43 -0
  46. sglang/srt/model_executor/model_runner.py +5 -0
  47. sglang/srt/models/arcee.py +532 -0
  48. sglang/srt/models/deepseek_v2.py +2 -0
  49. sglang/srt/models/glm4_moe.py +3 -1
  50. sglang/srt/models/granitemoe.py +3 -0
  51. sglang/srt/models/grok.py +3 -0
  52. sglang/srt/models/hunyuan.py +1 -0
  53. sglang/srt/models/llama4.py +3 -0
  54. sglang/srt/models/mixtral.py +3 -0
  55. sglang/srt/models/olmoe.py +3 -0
  56. sglang/srt/models/phimoe.py +1 -0
  57. sglang/srt/models/step3_vl.py +994 -0
  58. sglang/srt/multimodal/processors/base_processor.py +15 -16
  59. sglang/srt/multimodal/processors/step3_vl.py +515 -0
  60. sglang/srt/reasoning_parser.py +2 -1
  61. sglang/srt/server_args.py +10 -13
  62. sglang/srt/speculative/eagle_worker.py +2 -0
  63. sglang/utils.py +0 -11
  64. sglang/version.py +1 -1
  65. {sglang-0.4.9.post6.dist-info → sglang-0.4.10.dist-info}/METADATA +3 -4
  66. {sglang-0.4.9.post6.dist-info → sglang-0.4.10.dist-info}/RECORD +69 -56
  67. {sglang-0.4.9.post6.dist-info → sglang-0.4.10.dist-info}/WHEEL +0 -0
  68. {sglang-0.4.9.post6.dist-info → sglang-0.4.10.dist-info}/licenses/LICENSE +0 -0
  69. {sglang-0.4.9.post6.dist-info → sglang-0.4.10.dist-info}/top_level.txt +0 -0
@@ -210,6 +210,7 @@ class PhiMoE(nn.Module):
210
210
  self.experts = FusedMoE(
211
211
  num_experts=num_experts,
212
212
  top_k=top_k,
213
+ layer_id=layer_id,
213
214
  hidden_size=hidden_size,
214
215
  intermediate_size=intermediate_size,
215
216
  reduce_results=True,