sglang 0.4.0.post2__py3-none-any.whl → 0.4.1__py3-none-any.whl

{sglang-0.4.0.post2.dist-info → sglang-0.4.1.dist-info}/RECORD

@@ -1,17 +1,17 @@
 sglang/__init__.py,sha256=b2oIdWzp5P8SzieeOs2TzJoN3Do3tfJbV8gZS_imVcs,1619
 sglang/api.py,sha256=NdO6cYnklnEBQBKqQjlqI8-P1EownKQ71t5ibCGhEVo,6953
 sglang/bench_latency.py,sha256=oZjSAzX7dUiSu-zdz0dkyUPo-qAX_lsXFH1gf03akgI,76
-sglang/bench_offline_throughput.py,sha256=rgMWDhA1Hai0gKBzxc0dzTWfI8l39Cyw2VOCyMt1YyY,12771
-sglang/bench_one_batch.py,sha256=aF0onHeRjy7AYVjsq1IA3rZEhUuYXuslg1fAhuvJ2yo,16120
+sglang/bench_offline_throughput.py,sha256=iQiJCK3KQDCdwU1NVbIwbtthssWzBXiIsKUDA7Z_hO0,12510
+sglang/bench_one_batch.py,sha256=jkyMhK0lqn5dRCYgAh30qZrNHP4gAbXODymBMNXK86I,15859
 sglang/bench_one_batch_server.py,sha256=-fV9FTLNNcSIy0pgYeggXedPVK0fVsXZqVQswT8OMOY,5945
-sglang/bench_serving.py,sha256=zv_EcbWno79j7WYFL2m6BfCLT6iSOfGV4uwGbDg9KQA,53141
+sglang/bench_serving.py,sha256=3VQatM51v9f55aUQQ5crYMxxKHr1AbThicsWfBy_tjU,53190
 sglang/check_env.py,sha256=4OqpZaEJOfBM6-vtPILto5kqDmgiZM1Koc7lK78A7CI,8427
 sglang/global_config.py,sha256=fnT0U9vlHdGaQFKN9tYTnUF4-eVW4HYQURd5zvPtrg0,1286
 sglang/launch_server.py,sha256=4y2QeSj0wVNB9MJQZeahD4ahTDU6gwqo7MPUytyFop0,403
 sglang/launch_server_llavavid.py,sha256=tGc17S1vUfLwbi1GB26oOdXxTWr7gjlqpTrPnrMRNO8,1007
 sglang/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
 sglang/utils.py,sha256=23jf4Mz8E5p5a6JOkjnfYZixdjZUk88F_mZ8rZcby5Q,11597
-sglang/version.py,sha256=OUNovuQ1RrdJFYetl0e0U0556H_wiyjhVks9-l-zF94,28
+sglang/version.py,sha256=pMtTmSUht-XtbR_7Doz6bsQqopJJd8rZ8I8zy2HwwoA,22
 sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/lang/chat_template.py,sha256=cnfjjxIIcYRGRxXlJlOGnpFxFuhMHut7DS52LsOMKcA,15826
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
@@ -27,18 +27,19 @@ sglang/lang/backend/openai.py,sha256=qM7eVH_kMxnDd2rpxOH0v76KxtOJFlAwgLgWIKvFGCI
 sglang/lang/backend/runtime_endpoint.py,sha256=dfs-yZ1ekKmnbpZLluQHWPmMeZJKbaaZRRGYRa9eBE8,10541
 sglang/lang/backend/vertexai.py,sha256=O-iBLD-y3vq80UxnrAoJri7bxpgd-_eakZ88Cf8bEGA,4855
 sglang/srt/_custom_ops.py,sha256=Y4gyTDGhWz-W2Igq25Ojm8XFiyvkawW9I-79iwYvxJ0,3574
+sglang/srt/aio_rwlock.py,sha256=6LYtOdeTUY3hkfa1dmYkgsaF2ttrwIF3hUWz2AZ2fqw,2970
 sglang/srt/conversation.py,sha256=u9zFU8aMYzwHUbQRKU76B_T-jfLlPoxUcWG_nRbDM2I,21201
 sglang/srt/hf_transformers_utils.py,sha256=38Ms0H2-VMerOS6jnczcFtZMS6lhw9B5rSWKAfxVUfQ,7945
 sglang/srt/mm_utils.py,sha256=1ScBunw_x4W8ebM_AcJ62-1T2mfT8NlMJqdAhkF1lb0,12367
 sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,6078
-sglang/srt/server.py,sha256=tEciMH_U6WIZYPUGDDM0c4BQ16cvgVdA4II-ksPZoMo,34621
+sglang/srt/server.py,sha256=E9YKKXpXv3vPvRy0-cgcy0-5UA-OZz42-32EZWKTicA,34661
 sglang/srt/server_args.py,sha256=LgnQ-kBJZ3E7hMMZj9bSK0mn7Bhjk1nJHxLcxl-lGTM,34572
-sglang/srt/utils.py,sha256=WWEcMJHmvlOjiqE9UicT0ZYwa2PUKDZorAk2Y8PPRBI,42039
+sglang/srt/utils.py,sha256=J8kFl6kDBwFZCM6AKaVTiqdhJKRg0JOH0pNrD1ZeWmM,41726
 sglang/srt/configs/__init__.py,sha256=_usVIXHQjft4PAJ1Y-yGQOn2QNOv501GYMlQwpGXbns,208
 sglang/srt/configs/device_config.py,sha256=dResqHjkg_dq10v6rnVpbXpvABZRB0jylOm-2_JAnx0,428
 sglang/srt/configs/exaone.py,sha256=Duxd4yQoKy8GWEzZD_kCY_OzmN_67CTJL_Kgn0eXk3g,10731
 sglang/srt/configs/load_config.py,sha256=TcPi_HY6xu5SiVZsxPOoB5pGeDUNebOk7muoUH9VBDg,3083
-sglang/srt/configs/model_config.py,sha256=OjEeigs5tMNKP-RImJk2NHVFXv-fyQfsGREWMO3rqhM,15839
+sglang/srt/configs/model_config.py,sha256=vVarlLTw9Ged1PXIwRP-R8UhiG6oaezNIZhTNuF0eQc,16070
 sglang/srt/configs/qwen2vl.py,sha256=ZjLy9v2eZY4wptUfY3CWgYKg2B5DDrkfCSyTy_Zf_bg,4351
 sglang/srt/constrained/__init__.py,sha256=UWZNVLvOT5ZBX8M36sONgDmnKtkQ0cSfhQD2jO0ATuk,786
 sglang/srt/constrained/base_grammar_backend.py,sha256=FhVm7PxhXDl0joV9NP5RjKgz7dR1dZvUAQnh0mdtvVY,2353
@@ -60,35 +61,37 @@ sglang/srt/distributed/device_communicators/shm_broadcast.py,sha256=WVxBd1QfIgRW
 sglang/srt/distributed/device_communicators/xpu_communicator.py,sha256=P3WKgddcfpUhBa-_5PvjYxH146ZE-N1cotTzEpPRKlY,1620
 sglang/srt/layers/activation.py,sha256=EboMjT9HV2tNHQ6rzpojtlkzev1lAFbhQlxMg9hwxBQ,5471
 sglang/srt/layers/custom_op_util.py,sha256=0vu-yX2wwonmO1L_o5G7SA6C-8XuhDIh9rPDvNeLhoc,922
-sglang/srt/layers/fused_moe_patch.py,sha256=DMIyrwOON7OSidKZdreL5HzMhP0AD5Ues0xdY-ADOQw,4471
 sglang/srt/layers/layernorm.py,sha256=nRQ1w1xSUcU-zlqVC61BnGG6otS5W1w9VaSzeXizrx4,4037
-sglang/srt/layers/linear.py,sha256=dF2HvqiMbhWlCjvkLFRCcgUFGhG-B0keM_CIpjvgTtg,46154
+sglang/srt/layers/linear.py,sha256=KyRFU0VcoNuN-hnQB9QQcBN9NCpeqPtLzzufIHUpV6w,47064
 sglang/srt/layers/logits_processor.py,sha256=JlOU0x8vBGIuTwHSdjR6Kly9_uzilBMv0NE_rvUx0W4,14747
 sglang/srt/layers/pooler.py,sha256=rj2lygvleBnyLCBZ8I11HGMgpfIDsT0l3PIkshJwdu4,1606
 sglang/srt/layers/radix_attention.py,sha256=E4cmvkcCdCtb6VyLNrCKy1D6VwHQ063oH3JQXPaRy6w,2178
 sglang/srt/layers/rotary_embedding.py,sha256=29tx3JNR40AoXqBa2cFGBjva9vU2xgFipETlpMaaZas,3985
 sglang/srt/layers/sampler.py,sha256=k4Op_HMkQfT7t9wgQwBVotfTUXEocrzRyQqEFnff1pc,5511
-sglang/srt/layers/torchao_utils.py,sha256=07Fe2Csdh1JiQKPGGHWkbq0-a6bV7Cq136ygdtVAhgI,3708
+sglang/srt/layers/torchao_utils.py,sha256=dQVuWNXxAvOPjr2G5BBMWqC2oKcS2B52rx-fEc_elmc,3545
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=slGwLiWjuFLCUdRe-GTlfumyZpqVX9VF6No_UGOT-hA,21624
 sglang/srt/layers/attention/__init__.py,sha256=KIJhzOJWYioQE7Va4D83-V-ZUZVMZcczuNgDC3dlSRo,2583
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=RQdEKRykSLf9ilnaHmR6T7RFqh4emH_adfB3aJN2BUU,10920
-sglang/srt/layers/attention/flashinfer_backend.py,sha256=NgeigL1WiPOuOry0Gbxv-6HEcERB8Du0mBJgYcTVIAA,24943
+sglang/srt/layers/attention/flashinfer_backend.py,sha256=umD1E2zvMnPbbgvx2Ex5LQB6a4a41brjsks1M0gFMMU,26357
 sglang/srt/layers/attention/torch_native_backend.py,sha256=nQdeqWEMMH_wrod5wssDCJG-uPKm0uslvkALKqPRPQ8,10509
 sglang/srt/layers/attention/triton_backend.py,sha256=-TobyZHwlbJ5HhbFg-jgCqVOw4Y-opgEuFo-EusASQc,6264
 sglang/srt/layers/attention/triton_ops/decode_attention.py,sha256=oJ_UK1t229zF3hbTDiQe7t-X-IbM2dOxx4U2ch-vmjA,17847
 sglang/srt/layers/attention/triton_ops/double_sparsity_attention.py,sha256=1pSXfY3EEaM7iRN_uElHnAfsrJMhTFbu9fj8Z0O2PbE,21480
 sglang/srt/layers/attention/triton_ops/extend_attention.py,sha256=tZJhzqcf1KKMT8z7_32eVk_D1NHP71c-S3UNxemfAHM,11542
 sglang/srt/layers/attention/triton_ops/prefill_attention.py,sha256=lojFXRZMLWkzS2Y8uxaolnQhXaWKG19mCAWaF5KQeiI,6087
-sglang/srt/layers/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sglang/srt/layers/ep_moe/kernels.py,sha256=wb_S2qLxoWWgQu9coXy0XLNGvHzdZSdwXr0PGy4QySg,10940
-sglang/srt/layers/ep_moe/layer.py,sha256=uMropMhU-MaycoxSLxcfD0jZC_cuL_boRbIu86mbZjY,23034
-sglang/srt/layers/fused_moe_triton/__init__.py,sha256=PHKFqd2hPOO-g9kSMseg2g76lpg9OGXQDThWU6bt9vs,902
-sglang/srt/layers/fused_moe_triton/fused_moe.py,sha256=fLGmkY6imJYjEw9-3-jJthkMcFGMBcu9HCNIuxAzMhE,29625
-sglang/srt/layers/fused_moe_triton/layer.py,sha256=eMpbZlP3FAQxbHochis7ybZ-fsNBP0PzKF1PN0Xo7so,21517
-sglang/srt/layers/quantization/__init__.py,sha256=FgNy_zNWMWnq3lEGyCSyfLSQtcZtWlq99JilkmEDW7I,4594
+sglang/srt/layers/moe/fused_moe_native.py,sha256=8q-LFZMSCGLc2_Gltp2lH0gSb4A1WOuKQW3wo3rpj5g,1601
+sglang/srt/layers/moe/topk.py,sha256=YjIiFqMERvkChkwZUqTrL_xaQyzsYsZzVUe4PzAhRZI,6299
+sglang/srt/layers/moe/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sglang/srt/layers/moe/ep_moe/kernels.py,sha256=wb_S2qLxoWWgQu9coXy0XLNGvHzdZSdwXr0PGy4QySg,10940
+sglang/srt/layers/moe/ep_moe/layer.py,sha256=6iQU5ZjQ8IXGoQ8ZlBuJqyQxYTEem9vXI6rbVIWKlZw,22303
+sglang/srt/layers/moe/fused_moe_triton/__init__.py,sha256=h9yMFAL_bagUf-qBED8gSWdCOb7d8IdA-pE-L_nIg8E,842
+sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=GVugCKapd3CvgkvPQ_FmQplC12-grv3n1FRkLJc6WhY,30790
+sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=BclDj5JyCy-8Bfue4broL1-IG6a4dUyggE9WQLa06sg,20575
+sglang/srt/layers/quantization/__init__.py,sha256=VPYXShHvbvkOgVBlkIqic4RhdJ1y6EZ3r34T-nZMT1k,4606
 sglang/srt/layers/quantization/base_config.py,sha256=daK9p0aijMszLUm1W4Pc33FK87MdqYK1NoWFKif-j80,4599
-sglang/srt/layers/quantization/fp8.py,sha256=3oIUPaD0PBXQyTKr44I0YJ8XXDdwyoS_-ZA97XdSxXE,24143
-sglang/srt/layers/quantization/fp8_utils.py,sha256=eJDLLDu8ZbrbE3BfFIf89JlIMPOP-14DesbeVsajW0Q,1035
+sglang/srt/layers/quantization/fp8.py,sha256=wNnpXLroIl7D98mlfCiXZPE9hrP5ricHrXY1WZBzEEo,30810
+sglang/srt/layers/quantization/fp8_kernel.py,sha256=v4-7hCQFyuUSZmeJS_5VDCu6a1-EGWXQ088FdPTjO_0,8137
+sglang/srt/layers/quantization/fp8_utils.py,sha256=HBJBaNcln1NrLxzw0ppUjMd6w-ryuGDDHCYJq7mRQac,4035
 sglang/srt/lora/lora.py,sha256=-o2mBmUvoVpdkgdAkWTARN4kfyep3UNEJLcg6moh0SU,15056
 sglang/srt/lora/lora_config.py,sha256=a2fTQESlCbG1xLiBYy4ptZ6c0Burcqyg1_6V1XSok-Y,1506
 sglang/srt/lora/lora_manager.py,sha256=DHiqdl0_4wQ5PxZBZtlCpP14515mDV2_H9tzL3Rdss8,12886
@@ -96,23 +99,23 @@ sglang/srt/managers/data_parallel_controller.py,sha256=psI4FAuBGjtdnEuwagnGdtRqv
 sglang/srt/managers/detokenizer_manager.py,sha256=nZkbwt4yty_oy8rvg4T7PbgyVLoBLohvHl25xlQpBoo,8439
 sglang/srt/managers/image_processor.py,sha256=Y8RgyrzbJjJTpjbnZDa5qiiG5wWjZ68rOXUPDi6kkFo,13698
 sglang/srt/managers/io_struct.py,sha256=_LWWqT3LNwZGaWhg2d3kTg1V2MTHKzRasCvxF9Nfpi4,15429
-sglang/srt/managers/schedule_batch.py,sha256=SAd7sxhoC3Bp8_xd-TEcXEFZBlGZPbn8-wMvBcjU55Q,45607
+sglang/srt/managers/schedule_batch.py,sha256=qryPWCdOTFzxomDa80U-5guShOb1K4kBUWcPCCchYB8,45762
 sglang/srt/managers/schedule_policy.py,sha256=cLNi__smbg02keWgUMfB_nEM3vllocPB0XyG1P5qO7I,15469
-sglang/srt/managers/scheduler.py,sha256=QlcVMtrLlNcBOkVISdO556jrK8a4LE4ULskC0oCH2IQ,61776
+sglang/srt/managers/scheduler.py,sha256=3Olw4Yf4Qtn1i4PqK3PT9hkXYGE8nemL2_Xjn8JLxAQ,61819
 sglang/srt/managers/session_controller.py,sha256=Yp-IV3rXczACZxZXmF-QxW9CWICGy8KHQ9ttBGJ8WXA,2800
-sglang/srt/managers/tokenizer_manager.py,sha256=gnCCdB5XDobOoBKptwv-o0yYqDkMUxL78s0zBno5lM4,31219
+sglang/srt/managers/tokenizer_manager.py,sha256=Vta7Lysvh4rPWqEB00shqAzpGUfv7GdPETDqFCU8RxA,31556
 sglang/srt/managers/tp_worker.py,sha256=X1EwFX3FSsmXx7jeeX2tjZRocaujabQYWm-M-0CFEBE,7363
 sglang/srt/managers/tp_worker_overlap_thread.py,sha256=-QNBJRKxraa9Xt2WI1AFzZYdneIJ1eXv0GjFzDqXoE0,8926
 sglang/srt/mem_cache/base_prefix_cache.py,sha256=QC8HS8RC5DXu14kyXsxAgEUsn0f932p2DjqzbKjc6Bs,962
 sglang/srt/mem_cache/chunk_cache.py,sha256=R2gHAuqKd5ayQW3NnsgoGUH31---Z5izCDyCqLL0FjQ,2524
 sglang/srt/mem_cache/flush_cache.py,sha256=GYcxmNXh4hsMpFfNOuCTpKilW7guZwTtAg_usVeM3J0,979
-sglang/srt/mem_cache/memory_pool.py,sha256=l9_srwXEfIIDF46nxykbHIOo1VSvU5_Ew3H0r5EC7Fo,11072
+sglang/srt/mem_cache/memory_pool.py,sha256=oxk3UtiiFA3_1iIP6eFsk8HIcRI_8Z1-FE2KOWDr-YM,11366
 sglang/srt/mem_cache/radix_cache.py,sha256=c5voySV5L855c0G9cBEc9iQ4nR7PDDmg0V6fWWJHcq4,10945
 sglang/srt/metrics/collector.py,sha256=ZWoFx_FKN0sNMSZ8RJWUVQ0RFEYhIHxdw0d4TZTluMU,6861
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
-sglang/srt/model_executor/cuda_graph_runner.py,sha256=kZ3nV03MD8EQYQB38u4_88_wyW4unECxAdMVICpPyuk,16241
+sglang/srt/model_executor/cuda_graph_runner.py,sha256=1n5WxoE9-0B3unwkkcR355K_D290h2LGt_7EvH02DQM,16246
 sglang/srt/model_executor/forward_batch_info.py,sha256=L5mVoW5SaO6To-7nGk0TZM-FFB5_78cARpJ-aC2rwD0,12883
-sglang/srt/model_executor/model_runner.py,sha256=MLYBcYIQihu2I3PBTUghiU2mSWsDMzlKzcnX7yHa9JU,29837
+sglang/srt/model_executor/model_runner.py,sha256=Bm3NWTS3xmOGXEJnucnJZQldpVOzu-DCEUfaJy_PTU0,30104
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=VBrY4W9CiVvS_D8yXhdkW9jReV9rSMSkJplabz0Fxgk,43528
 sglang/srt/model_loader/utils.py,sha256=0NaMR67fESFopaklmsleiL27XH1QUrjZW246MUu1EJ0,1369
@@ -120,9 +123,9 @@ sglang/srt/model_loader/weight_utils.py,sha256=kQo9KPThjH3HAOCfC_tdwdrshdWuWJOVp
 sglang/srt/models/baichuan.py,sha256=PzBOFcEAixakPEkQSaJwC0Xc1fu-yCsN9T0I67r8QmY,14919
 sglang/srt/models/chatglm.py,sha256=DOrEhmb0s-yPId88R6nJeLOTUEtogk-vkB69qT2JdWc,12913
 sglang/srt/models/commandr.py,sha256=PNXgfOZF84h-rSH0edEECUmEGW8YLb44V75Z_oDhFiA,14223
-sglang/srt/models/dbrx.py,sha256=2Wqcf3sv57l4gi2xH8yrb5WSmY-4_kbbf6fhpJ4aKWw,14581
-sglang/srt/models/deepseek.py,sha256=BVNICGoLjQoHmR5lc31YrZ6YbxSRTBilHqlLsALr2u8,15693
-sglang/srt/models/deepseek_v2.py,sha256=YKSrqagVcSUwCAi-rwIph-Xu12GrNETMNKxgnffWod8,35349
+sglang/srt/models/dbrx.py,sha256=okIpIwdr8Cfrz_thzc1F75XqCUfHhFLvZ1B6BaswKoA,14585
+sglang/srt/models/deepseek.py,sha256=_cVOvR6eSEgRf6TUBpTD5uMdijDWFw4sSt4lGzl8tbg,15697
+sglang/srt/models/deepseek_v2.py,sha256=-v_OJr2c3gJ0NMxQjvT3Jknz1XPGkzKx0TVR3NIiC6A,37284
 sglang/srt/models/exaone.py,sha256=dkERTZVxrRroqu5AGLP7D4N6n8HvDqlNaDQUIe15mZY,13038
 sglang/srt/models/gemma.py,sha256=ydRqsG-7004r1fAiz01LHUmcj_6XN0Tn4xO1keJnMQk,12126
 sglang/srt/models/gemma2.py,sha256=41PlW8pMb4rMETdAni_JWDhZeIn_QsTQireAyUjsURA,15848
@@ -130,7 +133,7 @@ sglang/srt/models/gemma2_reward.py,sha256=nJ01KfqLSJtqMLm3sG8p2mGZFK1xhhjh7I7Ccb
 sglang/srt/models/gpt2.py,sha256=2je1kE09sGcaORWnJuGYAkcwwOrT9EK-KhQaoCKjCSA,9517
 sglang/srt/models/gpt_bigcode.py,sha256=tovyOdJu2x3LkzmkdFXX_iJdkxuyChIDxwgvPBy6UPo,9528
 sglang/srt/models/granite.py,sha256=AeQY9Dxd1ZnwgCYBK0vSXXiMGM-yt9iaOVf_ruOUHXw,20409
-sglang/srt/models/grok.py,sha256=UWvVEYfEoH0jGNFSbXpO66OGW5pzmIHlNKcn9gRZEoQ,15664
+sglang/srt/models/grok.py,sha256=J9lgNbFebvXgF19nfZyHwlGPlGWY_m0LgP506YvOYrU,15668
 sglang/srt/models/internlm2.py,sha256=_xcKtd6YtEFUTozaN-yUb0xbSYckRpomfPSKcAk4j-Y,12127
 sglang/srt/models/internlm2_reward.py,sha256=8K26A9oIFFGx_9U2mF87j7FX8K87HGKMnVL3ht1Uc7I,2398
 sglang/srt/models/llama.py,sha256=S7nS05hhFGghXu0v-w9RZyBTY6OCEVF5Aaw4GX_E_9g,19929
@@ -142,24 +145,24 @@ sglang/srt/models/llavavid.py,sha256=dYUkKfHoE15vF_VXA_s_ICCTUMSmSgvP181fk8dUi0g
 sglang/srt/models/minicpm.py,sha256=ws4AqhOfAvYHGd04QuXCZel-Oxy9_vN4p4rTjs9RSz0,13723
 sglang/srt/models/minicpm3.py,sha256=YIKJDTpwjmpLlv1sNT93k2yZMvGQlI_H87czjf6QYyo,24707
 sglang/srt/models/mistral.py,sha256=EYifJUUzN2Z2-iL37eJiNZF_DB0H4pa0mKlgYRIxM70,838
-sglang/srt/models/mixtral.py,sha256=vi6ssY75kNLy_kJrDru6gJYiAogHjSniaO6aMFd1w4E,14515
+sglang/srt/models/mixtral.py,sha256=L2Gz-Cmih1V75Ks9jmI2a6rUQ1Cl6F2uDgrhDjjDJzs,14523
 sglang/srt/models/mixtral_quant.py,sha256=uuVO1nWUZJiDhbqZN6gzSMwyfpyZorMuFXHeMCGo7N0,14022
 sglang/srt/models/mllama.py,sha256=3kX-UqeTSYZL5kPNdkfKEAEv3DpSAW1ArAAoeiXVzIc,37739
 sglang/srt/models/olmo.py,sha256=OCDMtX1OI83r80mzU4FMC3Tg8cleQ-7C8Tpoe8zgzss,11708
 sglang/srt/models/olmo2.py,sha256=aC7svioN7XT5owRxPrvhvWBNMON9QXGQBWJ1KHMyXeA,13442
-sglang/srt/models/olmoe.py,sha256=Rw-3YrHWd90MZQFnmcfUQ-3wAaI0PCFKb0DIrCDND3s,15347
+sglang/srt/models/olmoe.py,sha256=LiHVGfRaC5c_BU_vVgtV9uLuDH_SC0dw1kEc61posmI,15351
 sglang/srt/models/phi3_small.py,sha256=44_my3QmgJ2N7SOkGZzEb62DXBeCVHojfmCWgkk2uCI,14802
 sglang/srt/models/qwen.py,sha256=_FKDbwaS5C07uJyyivZpBrXJVej4Ph9ivzJdzWJPxJ4,9904
 sglang/srt/models/qwen2.py,sha256=be4xgcuqNa9kBdaL7x3PjsnUky6fh5K33c_khAWSi04,12959
-sglang/srt/models/qwen2_moe.py,sha256=rYUk_vZW3ftKIIlqPvJZ1K-6oZ_PfGspixh1zm2Y8C8,16538
+sglang/srt/models/qwen2_moe.py,sha256=6xRRJxWWh1M5UFPfvhsCpY477zv-30AeSRJXsvOkgFc,16542
 sglang/srt/models/qwen2_vl.py,sha256=3EaUlTbyWOTRXA7eViK1WqmVbCFhXLIpnos49zzf-yM,26561
 sglang/srt/models/registry.py,sha256=inKh9iwOp3LFYm3nqujg-OtABClOP-ifc1stA9cZegA,3434
 sglang/srt/models/stablelm.py,sha256=iBlIkM7CQmqI25nsujWk0LLCQD7TshzUU8qzZYYrt20,11311
 sglang/srt/models/torch_native_llama.py,sha256=YeXHorFm6QfnczLXwPb5TG9a-He0uiA9RzpR1YZKGg4,18758
 sglang/srt/models/xverse.py,sha256=Oq--KqvbYu2H4TMVGEHpSnJLEwXBpxlncR9ilsQeckc,13579
-sglang/srt/models/xverse_moe.py,sha256=AawKEQw--oAl-yzwCjoaZRG7q3rdkyDiam3FS0zjf_c,15537
+sglang/srt/models/xverse_moe.py,sha256=7E60YIST4ELYwLRgjtHiLRI5Uyc7XqQTM7jQXiWaQs4,15541
 sglang/srt/models/yivl.py,sha256=88OubtuZ38Dxb2LzfV_MTPBI4wKhh4NJqFu--efbhFM,4809
-sglang/srt/openai_api/adapter.py,sha256=dvKq4O3Rhd77ad6iCtPNykgnk9PVJE-E8wHVsBAfCQQ,53927
+sglang/srt/openai_api/adapter.py,sha256=DbLA4-v-QrKJHYDH4fpDSXqmyz_vpcFE-1tnhh60m6o,54057
 sglang/srt/openai_api/protocol.py,sha256=ecRNNqkhwwKZaIoJlPhtp2VTcHxBJDbNN8lrKS7uBx8,10406
 sglang/srt/sampling/sampling_batch_info.py,sha256=s--zNjk-LErZ5lMqnZ7KiuJltaziKRbQAU5qYpKIxAc,8564
 sglang/srt/sampling/sampling_params.py,sha256=n7RbBg_bS5fYhsiWa8uJYnfoXy_i5DvtTBOkuFnHDNU,5286
@@ -180,12 +183,13 @@ sglang/test/simple_eval_math.py,sha256=6kGKNwNbLN-Af3Wj8WTimWhH-Xp3enDmSvvSjsgWU
 sglang/test/simple_eval_mgsm.py,sha256=rd7TSUyxdKbrXaVoewo24V8lCo_6kO8zxPhhmvylpw8,10259
 sglang/test/simple_eval_mmlu.py,sha256=FkwamjGMjueTixymkedF-YiPloSLiy4ftILFUrKZ9XI,4357
 sglang/test/test_activation.py,sha256=jkdNRzJnbd5OgZliQaIXpxovlcky17UrweomcOcMxoE,1442
+sglang/test/test_block_fp8.py,sha256=rhrIun8aW5zq2qvuGRlo7F7aZ_upjVxtQMVlyc2Th_E,11771
 sglang/test/test_layernorm.py,sha256=IacByD5d-stXjzBz8Ypamc7povlcedpKPbb_4JLgo3c,3720
 sglang/test/test_programs.py,sha256=1Z0umrsUu9pagzyGH5SrXl_qhKSyTfUv_kWC2mcn0qo,18208
 sglang/test/test_utils.py,sha256=HJG7kUQOk6n9FBbH89PDtQ41C3kt1cfJODhAEcFT0AQ,23823
 sglang/test/srt/sampling/penaltylib/utils.py,sha256=CjxHgywh0hx_87iynzQt_ztHu6zBVuE-YrZ-XPmW6U4,12906
-sglang-0.4.0.post2.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.0.post2.dist-info/METADATA,sha256=maHXecD3U1DdhzfU2aBMhN96MQRqCBPsIA1KlO7t7dg,22512
-sglang-0.4.0.post2.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
-sglang-0.4.0.post2.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.0.post2.dist-info/RECORD,,
+sglang-0.4.1.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.1.dist-info/METADATA,sha256=RlVEQtwr_CCGTs83vNPwWXQukutbFfBz9xBPlXSl6qc,22523
+sglang-0.4.1.dist-info/WHEEL,sha256=PZUExdf71Ui_so67QXpySuHtCi3-J3wvF4ORK6k_S8U,91
+sglang-0.4.1.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.1.dist-info/RECORD,,

sglang/srt/layers/fused_moe_patch.py

@@ -1,133 +0,0 @@
-"""
-Torch-native implementation for FusedMoE. This is used for torch.compile.
-It is based on https://github.com/pytorch-labs/gpt-fast/blob/32971d3129541c5bfb4f715abc33d1c5f408d204/mixtral-moe/model.py#L204
-"""
-
-from typing import Callable, Optional
-
-import torch
-from torch.nn import functional as F
-
-
-def fused_topk_native(
-    hidden_states: torch.Tensor,
-    gating_output: torch.Tensor,
-    topk: int,
-    renormalize: bool,
-):
-    assert hidden_states.shape[0] == gating_output.shape[0], "Number of tokens mismatch"
-    M, _ = hidden_states.shape
-    topk_weights = torch.empty(
-        M, topk, dtype=torch.float32, device=hidden_states.device
-    )
-    topk_ids = torch.empty(M, topk, dtype=torch.int32, device=hidden_states.device)
-    topk_weights = F.softmax(gating_output.float(), dim=-1)
-    topk_weights, topk_ids = torch.topk(topk_weights, topk, dim=-1)
-    if renormalize:
-        topk_weights = topk_weights / topk_weights.sum(dim=-1, keepdim=True)
-    return topk_weights, topk_ids
-
-
-# This is used by the Deepseek-V2 model
-def grouped_topk(
-    hidden_states: torch.Tensor,
-    gating_output: torch.Tensor,
-    topk: int,
-    renormalize: bool,
-    num_expert_group: int = 0,
-    topk_group: int = 0,
-):
-
-    assert hidden_states.shape[0] == gating_output.shape[0], "Number of tokens mismatch"
-
-    scores = torch.softmax(gating_output, dim=-1)
-    num_token = scores.shape[0]
-    group_scores = (
-        scores.view(num_token, num_expert_group, -1).max(dim=-1).values
-    )  # [n, n_group]
-    group_idx = torch.topk(group_scores, k=topk_group, dim=-1, sorted=False)[
-        1
-    ]  # [n, top_k_group]
-    group_mask = torch.zeros_like(group_scores)  # [n, n_group]
-    group_mask.scatter_(1, group_idx, 1)  # [n, n_group]
-    score_mask = (
-        group_mask.unsqueeze(-1)
-        .expand(num_token, num_expert_group, scores.shape[-1] // num_expert_group)
-        .reshape(num_token, -1)
-    )  # [n, e]
-    tmp_scores = scores.masked_fill(~score_mask.bool(), 0.0)  # [n, e]
-    topk_weights, topk_ids = torch.topk(tmp_scores, k=topk, dim=-1, sorted=False)
-
-    if renormalize:
-        topk_weights = topk_weights / topk_weights.sum(dim=-1, keepdim=True)
-    return topk_weights, topk_ids
-
-
-def select_experts_native(
-    hidden_states: torch.Tensor,
-    router_logits: torch.Tensor,
-    top_k: int,
-    use_grouped_topk: bool,
-    renormalize: bool,
-    topk_group: Optional[int] = None,
-    num_expert_group: Optional[int] = None,
-):
-    # DeekSeekv2 uses grouped_top_k
-    if use_grouped_topk:
-        assert topk_group is not None
-        assert num_expert_group is not None
-        topk_weights, topk_ids = grouped_topk(
-            hidden_states=hidden_states,
-            gating_output=router_logits,
-            topk=top_k,
-            renormalize=renormalize,
-            num_expert_group=num_expert_group,
-            topk_group=topk_group,
-        )
-    else:
-        topk_weights, topk_ids = fused_topk_native(
-            hidden_states=hidden_states,
-            gating_output=router_logits,
-            topk=top_k,
-            renormalize=renormalize,
-        )
-    return topk_weights, topk_ids
-
-
-def fused_moe_forward_native(
-    layer: torch.nn.Module,
-    x: torch.Tensor,
-    use_grouped_topk: bool,
-    top_k: int,
-    router_logits: torch.Tensor,
-    renormalize: bool,
-    topk_group: Optional[int] = None,
-    num_expert_group: Optional[int] = None,
-    custom_routing_function: Optional[Callable] = None,
-) -> torch.Tensor:
-
-    if use_grouped_topk:
-        assert num_expert_group is not None and topk_group is not None
-        topk_weights, topk_ids = grouped_topk(
-            x,
-            router_logits,
-            top_k,
-            renormalize,
-            num_expert_group,
-            topk_group,
-        )
-    elif custom_routing_function is None:
-        topk_weights, topk_ids = fused_topk_native(x, router_logits, top_k, renormalize)
-    else:
-        topk_weights, topk_ids = custom_routing_function(
-            x, router_logits, top_k, renormalize
-        )
-
-    w13_weights = layer.w13_weight[topk_ids]
-    w1_weights, w3_weights = torch.chunk(w13_weights, 2, dim=2)
-    w2_weights = layer.w2_weight[topk_ids]
-    x1 = torch.einsum("ti,taoi -> tao", x, w1_weights)
-    x1 = F.silu(x1)
-    x3 = torch.einsum("ti, taoi -> tao", x, w3_weights)
-    expert_outs = torch.einsum("tao, taio -> tai", (x1 * x3), w2_weights)
-    return torch.einsum("tai,ta -> ti", expert_outs, topk_weights.to(expert_outs.dtype))