sglang 0.5.1.post1__py3-none-any.whl → 0.5.1.post3__py3-none-any.whl

sglang/test/test_cutlass_moe.py

@@ -8,6 +8,15 @@ from transformers import AutoConfig
 
 from sglang.srt.layers.moe.cutlass_moe import cutlass_fused_experts_fp8
 from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_experts
+from sglang.srt.layers.moe.moe_runner.base import MoeRunnerConfig
+
+
+# Copy from: https://github.com/deepseek-ai/DeepGEMM/blob/main/deep_gemm/utils.py
+def calc_diff(x, y):
+    x, y = x.double(), y.double()
+    denominator = (x * x + y * y).sum()
+    sim = 2 * (x * y).sum() / denominator
+    return 1 - sim
 
 
 def get_model_config(tp_size: int):
@@ -69,16 +78,11 @@ def run_test(tp_size, batch_size, model_config, check=False):
 
     # --- Input Data ---
     # Use bf16/fp16 for input activation based on model config
-    x = torch.randn((batch_size, H), device="cuda", dtype=dtype) * 0.0001
+    x = torch.randn((batch_size, H), device="cuda", dtype=dtype)
     # --- Weights (Generate in higher precision, then convert to FP8) ---
     # Generate weights suitable for FP8 conversion (e.g., scaled appropriately)
-    w1_hp = (
-        torch.randn((E, I, H), device="cuda", dtype=torch.float32) * 0.00001 + 0.00001
-    )
-    w2_hp = (
-        torch.randn((E, H, I // 2), device="cuda", dtype=torch.float32) * 0.00001
-        + 0.00001
-    )
+    w1_hp = torch.randn((E, I, H), device="cuda", dtype=torch.float32)
+    w2_hp = torch.randn((E, H, I // 2), device="cuda", dtype=torch.float32)
 
     w1 = to_fp8(w1_hp)
     w2 = to_fp8(w2_hp)
@@ -149,13 +153,13 @@ def run_test(tp_size, batch_size, model_config, check=False):
     )
 
     # Note: Triton expects non-transposed weights
+    moe_config = MoeRunnerConfig(inplace=False)
     triton_lambda = lambda: fused_experts(
         x,
         w1,
         w2,
         (topk_weights, topk_ids, "dummy"),
-        inplace=False,
-        activation="silu",  # Assuming SiLU activation common in MoEs
+        moe_config,
         use_fp8_w8a8=True,
         w1_scale=w1_scale,
         w2_scale=w2_scale,
@@ -221,32 +225,19 @@ def run_test(tp_size, batch_size, model_config, check=False):
                 w1,  # Original shape
                 w2,  # Original shape
                 (topk_weights, topk_ids, "dummy"),
-                inplace=False,  # Important: Use False to get output tensor
-                activation="silu",
+                moe_config,
                 use_fp8_w8a8=True,
                 w1_scale=w1_scale,
                 w2_scale=w2_scale,
                 block_shape=block_shape,
             )
 
-        # Ensure outputs are same dtype for comparison
-        y_cutlass = y_cutlass.to(dtype)
-        y_triton = y_triton.to(dtype)
-
-        abs_error = torch.abs(y_cutlass - y_triton)
-        rel_error = abs_error / torch.clamp(torch.abs(y_triton), min=1e-2)
-
-        max_abs_err = abs_error.max().item()
-        max_rel_err = rel_error.max().item()
-
-        print("y_cutlass:", y_cutlass[:, :10])
-        print("y_triton:", y_triton[:, :10])
-        print(f"Max absolute error: {max_abs_err:.6f}")
-        print(f"Max relative error: {max_rel_err:.6f}")
+        diff = calc_diff(y_cutlass, y_triton)
+        print(f"Diff: {diff:.6f}")
 
         # Tolerance might need adjustment based on FP8 specifics and kernel differences
         # FP8 comparisons often require higher tolerance than FP16/BF16
-        assert max_rel_err < 5e-1, f"Relative error too high! {max_rel_err}"
+        assert diff < 1e-4, f"Diff too high! {diff}"
         print("Correctness check passed.")
 
 
@@ -264,7 +255,21 @@ if __name__ == "__main__":
         "--batch-sizes",
         type=int,
         nargs="+",
-        default=[1, 4, 8, 16, 32, 64, 128, 256, 512, 1024],  # Adjusted default
+        default=[
+            1,
+            4,
+            8,
+            16,
+            32,
+            64,
+            128,
+            256,
+            512,
+            1024,
+            2048,
+            4096,
+            8192,
+        ],  # Adjusted default
         help="List of batch sizes to test",
     )
     parser.add_argument("--check", action="store_true", help="Enable check mode")

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.5.1.post1"
+__version__ = "0.5.1.post3"

{sglang-0.5.1.post1.dist-info → sglang-0.5.1.post3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sglang
-Version: 0.5.1.post1
+Version: 0.5.1.post3
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -257,12 +257,12 @@ Requires-Dist: uvloop; extra == "runtime-common"
 Requires-Dist: xgrammar==0.1.23; extra == "runtime-common"
 Provides-Extra: srt
 Requires-Dist: sglang[runtime_common]; extra == "srt"
-Requires-Dist: sgl-kernel==0.3.5; extra == "srt"
+Requires-Dist: sgl-kernel==0.3.7; extra == "srt"
 Requires-Dist: torch==2.8.0; extra == "srt"
 Requires-Dist: torchaudio==2.8.0; extra == "srt"
 Requires-Dist: torchvision; extra == "srt"
 Requires-Dist: cuda-python; extra == "srt"
-Requires-Dist: flashinfer_python==0.2.11.post3; extra == "srt"
+Requires-Dist: flashinfer_python==0.2.14.post1; extra == "srt"
 Provides-Extra: blackwell
 Requires-Dist: sglang[runtime_common]; extra == "blackwell"
 Requires-Dist: sgl-kernel; extra == "blackwell"
@@ -270,7 +270,7 @@ Requires-Dist: torch==2.8.0; extra == "blackwell"
 Requires-Dist: torchaudio==2.8.0; extra == "blackwell"
 Requires-Dist: torchvision; extra == "blackwell"
 Requires-Dist: cuda-python; extra == "blackwell"
-Requires-Dist: flashinfer_python==0.2.11.post3; extra == "blackwell"
+Requires-Dist: flashinfer_python==0.2.14.post1; extra == "blackwell"
 Provides-Extra: srt-hip
 Requires-Dist: sglang[runtime_common]; extra == "srt-hip"
 Requires-Dist: torch; extra == "srt-hip"
@@ -304,6 +304,7 @@ Requires-Dist: pandas; extra == "test"
 Requires-Dist: peft; extra == "test"
 Requires-Dist: sentence_transformers; extra == "test"
 Requires-Dist: pytest; extra == "test"
+Requires-Dist: tabulate; extra == "test"
 Provides-Extra: all
 Requires-Dist: sglang[srt]; extra == "all"
 Requires-Dist: sglang[openai]; extra == "all"
@@ -374,7 +375,7 @@ Dynamic: license-file
 | [**Slides**](https://github.com/sgl-project/sgl-learning-materials?tab=readme-ov-file#slides) |
 
 ## News
-- [2025/08] 🔔 SGLang x AMD SF Meetup on 8/22: Hands-on GPU workshop, tech talks by AMD/xAI/SGLang, and networking ([Roadmap](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_sglang_roadmap.pdf), [Large-scale EP](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_sglang_ep.pdf)).
+- [2025/08] 🔔 SGLang x AMD SF Meetup on 8/22: Hands-on GPU workshop, tech talks by AMD/xAI/SGLang, and networking ([Roadmap](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_sglang_roadmap.pdf), [Large-scale EP](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_sglang_ep.pdf), [Highlights](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_highlights.pdf), [AITER/MoRI](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_aiter_mori.pdf), [Wave](https://github.com/sgl-project/sgl-learning-materials/blob/main/slides/amd_meetup_wave.pdf)).
 - [2025/08] 🔥 SGLang provides day-0 support for OpenAI gpt-oss model ([instructions](https://github.com/sgl-project/sglang/issues/8833))
 - [2025/06] 🔥 SGLang, the high-performance serving infrastructure powering trillions of tokens daily, has been awarded the third batch of the Open Source AI Grant by a16z ([a16z blog](https://a16z.com/advancing-open-source-ai-through-benchmarks-and-bold-experimentation/)).
 - [2025/06] 🔥 Deploying DeepSeek on GB200 NVL72 with PD and Large Scale EP (Part I): 2.7x Higher Decoding Throughput ([blog](https://lmsys.org/blog/2025-06-16-gb200-part-1/)).

{sglang-0.5.1.post1.dist-info → sglang-0.5.1.post3.dist-info}/RECORD

@@ -1,15 +1,15 @@
 sglang/__init__.py,sha256=CK60u1J439CJZaPihWqlDjFZ8doIC-5YgIYz-E1PIh4,1823
 sglang/bench_offline_throughput.py,sha256=Bge0KG4Uiv9CHsxDnqKdXJmG_TcxM4blz3CmvLmDRvM,14815
 sglang/bench_one_batch.py,sha256=8Aic1FibvTpRtxtZx0fK97v8n7MQYLzg9Hg4t64iorQ,23041
-sglang/bench_one_batch_server.py,sha256=c9n1gsGI435eQz9dOqKAP11t32kmkCLJ_EW4gWtPSeU,14340
-sglang/bench_serving.py,sha256=xR1blR5t0t6YDEIOXvxjhrL7XkptBBV4VeQGZIORvLQ,71791
+sglang/bench_one_batch_server.py,sha256=d-XnTgyUzsEeOHyuc-DGZDeW4ZB4V59aV9Ojr3H43SE,14676
+sglang/bench_serving.py,sha256=TcT-yEO9bJW4VASIrfJaDZwKepBcerBUmVPeKbWoruI,78205
 sglang/check_env.py,sha256=qDMIG2rCNBH1yKnxQmF-Bp10oiFMUKMgfZLHZYOmdSY,8412
 sglang/compile_deep_gemm.py,sha256=EsgboGA-MK3Rvx9TlUJPLxJ5LBira4bTcHJLKEp1H6k,6488
 sglang/global_config.py,sha256=ZMTux_PsGnvkyJ0kTFwhTdbnFwIjjpGDogut_9Lu4Vo,1732
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
-sglang/profiler.py,sha256=tEHzHerXC-ymk4OrkoUcMbgcGHmb8VESthsNSP2Yx9w,4417
+sglang/profiler.py,sha256=JCpZzlDhahoiPlPi5IG3n7GFrQHxfHEB6ELie3Ck55w,4397
 sglang/utils.py,sha256=dC2PNkKYTgDHsNrWdZJ74GvaXGSHCeIk_aZ-TA89OhY,16380
-sglang/version.py,sha256=cOaLNeTwkPXJFU_tSZi-530rXF3qjxypENpk5W_HqGE,28
+sglang/version.py,sha256=_jbt9dYNR7pmoDsN781ZGkqYq_dZ7gNo3S1g-kmR8DM,28
 sglang/eval/llama3_eval.py,sha256=mLNRZJIqV4CfqrY8UGnJEcHw2Xsyr1eyYZgFSUFYr1g,9997
 sglang/eval/loogle_eval.py,sha256=-CC2s2kh5qUoDrHRkQVkC_jNvBgNojXbf456ny5s78s,4557
 sglang/lang/api.py,sha256=rcp3GeoyZhmJ0GDLPRkuZNcxd0TBJy_wfUDpcmQoqW8,7210
@@ -30,22 +30,23 @@ sglang/srt/aio_rwlock.py,sha256=6LYtOdeTUY3hkfa1dmYkgsaF2ttrwIF3hUWz2AZ2fqw,2970
 sglang/srt/bench_utils.py,sha256=zNTsTBUta27XxojgvR5p5X9M0v1MFzZ7dcX9TMq4NvQ,4348
 sglang/srt/code_completion_parser.py,sha256=KFa95OU0TeVjJkOqIgS4xV3kaJ5dFWOmAAgISyc1oEc,3803
 sglang/srt/constants.py,sha256=0i-tEwG2BSYNDy96MxnGHV5HnBELkYcnsVGsE-R18o0,93
-sglang/srt/conversation.py,sha256=7WkS3GZjGVt82hUglXdYBUdf13vk6U3-mN6KAGPMrCw,37822
+sglang/srt/conversation.py,sha256=cmVT4SCqRYxprYvmftUC_z_TkhXCqYC9B2rVZsmQe30,38869
 sglang/srt/custom_op.py,sha256=7OPQAZe3PqDXk-xM0Wxl7pm9cazilNJfkNiYIWztWgI,3198
-sglang/srt/hf_transformers_utils.py,sha256=WgpIr56YuKG8Hmp_vrZBfePYWMzDPVhH8k0p4DLk5sA,14877
+sglang/srt/harmony_parser.py,sha256=YweG4azWm5jmzYGaAHEXSaORpgWctPpSGQk_jp5H-CE,23227
+sglang/srt/hf_transformers_utils.py,sha256=YXaapaY6N8ETS0vTQBwcFERG5eMt3nsdN6GYeiEK9wM,15201
 sglang/srt/host_shared_memory.py,sha256=00TgVm3PNCl4rTPJZChG5uu4I5hd_Xf587r39tibmIg,2274
 sglang/srt/jinja_template_utils.py,sha256=XCuqi6iHQrtziUvhy22ePdh8WcS42O9sKxuXpM_bMXc,7836
 sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,6078
-sglang/srt/offloader.py,sha256=Bgob16HwmiQXt5yV0okin-UgcdMLHpEDXI0OtHfi6vI,14138
+sglang/srt/offloader.py,sha256=iu6Q5lOzJvuQVmq_CEYWitu41sNF1wmavTvOxlyEGaU,18097
 sglang/srt/operations.py,sha256=za_h1_kcR1Te6-i_r0i_zAeumbmIRhLK2pS31sukkrw,6089
 sglang/srt/operations_strategy.py,sha256=-w6GI2VgSGvORy3pUWRAJo6244ykSW1ZBV1Dx0fZlHA,7138
 sglang/srt/patch_torch.py,sha256=OUPCGQSQz3MVZB1zZ_Eq8lXiw0uIKJ_HWjqQolI8FsM,3088
 sglang/srt/poll_based_barrier.py,sha256=9vjn32ROlSueEk7EuZjWKUrrL7XVR9KJsSVliGTYB_4,937
-sglang/srt/reasoning_parser.py,sha256=er9PSDKroGhXG52k0Kz9Aw5o_7rpN1LXVp5mmZgtzqY,22948
-sglang/srt/server_args.py,sha256=-HMb560gDy7a3_HDh2St8nvkLv00dyKCX2s22ufkNew,102455
+sglang/srt/reasoning_parser.py,sha256=HEWAeFzPA_Jn3a44BYCz61QNV6kAvX46Y0tR8csAUg4,11152
+sglang/srt/server_args.py,sha256=ywi3C49ixZneuJ7EXjTu4E1EMn0M6-jlQfHfU77ksLQ,102532
 sglang/srt/torch_memory_saver_adapter.py,sha256=K_eTx0UU84MHSTXI3iqYLdHV4IWtJMJ2FKdGFJR8v1E,2417
 sglang/srt/two_batch_overlap.py,sha256=UykF5nC2rja3Hvmu0D9glqKdVRIEhQGPV84Jm7veopQ,34150
-sglang/srt/utils.py,sha256=ts1LtMbDygqKdlDunDmMn7wIk_6_X3WTZ0U_tzs4eXQ,93412
+sglang/srt/utils.py,sha256=hgJ3PsNL07eOe4uvIdq50_Aiq2WJo0783FSJJ_L9xwk,95170
 sglang/srt/warmup.py,sha256=zldxhMlXpclRAJXmfBjJNUJd1eDizVdysibBvQyTVuA,1782
 sglang/srt/configs/__init__.py,sha256=gas3cKYVgZp7qF4m8FejX4EUUmdQd_w5WDTm7fo7yM0,743
 sglang/srt/configs/chatglm.py,sha256=j-b0YkdYUmQm2y1kNmMJtKeACxWKmBbvNNkDWbs6kbI,2907
@@ -81,12 +82,12 @@ sglang/srt/debug_utils/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3
 sglang/srt/debug_utils/dump_comparator.py,sha256=8ZPwP0zojfRHvPiSKgpESYFcC3tR-sBjIej5SotOmpg,3947
 sglang/srt/debug_utils/dumper.py,sha256=eyrg5lmwyYF88kn8b06kpu1__aGnWb6CLM8fxywtooI,3158
 sglang/srt/debug_utils/text_comparator.py,sha256=pg58q3fzsnd3rJKRgZTdcOviOtAv8pSYZdwf-TLffMM,5522
-sglang/srt/disaggregation/decode.py,sha256=sdV8sGhwk4sX-V4p3xjxkqql0YZjsgnPOf6dG5UJSB4,34325
+sglang/srt/disaggregation/decode.py,sha256=EELA3krHp4YtRIpUYU0-9LSFd7XQuvKn3yiHGrK7gaI,34598
 sglang/srt/disaggregation/decode_schedule_batch_mixin.py,sha256=Iu7ehHGAfYIVATtClwX6y-6zxWs-JyDlMotYo_5yuUc,6598
 sglang/srt/disaggregation/kv_events.py,sha256=yFbtwOeblcCmOXTjg00TopxnyjkmCBQIVz46KB8jetY,13555
 sglang/srt/disaggregation/launch_lb.py,sha256=pmHwo1GWbpHZn4IDKnOEHYVcfdKpsYuTHUMowNo3TrE,3891
 sglang/srt/disaggregation/mini_lb.py,sha256=4l3W2COZuX8DbLg8AQvAhv5EncH9FkEv2HF5fhn3GQM,14630
-sglang/srt/disaggregation/prefill.py,sha256=wA-sXQG0t3hqzXDZcp5H6wI5F6qkZ-3CyzIX5Nc5v_w,36019
+sglang/srt/disaggregation/prefill.py,sha256=CLoixMP1qRWO-VEsvVbjfQOu4ggqF0KeuHhxZC-NSvU,36272
 sglang/srt/disaggregation/utils.py,sha256=evdv4AkQxJzTqdO0jrdAr7vHUbBSCEXMzNREh0FoRLc,12294
 sglang/srt/disaggregation/ascend/__init__.py,sha256=-lxnica6fZYmYNpKwmE8yESUpQ5Sxf2DiZoBHzboKc8,146
 sglang/srt/disaggregation/ascend/conn.py,sha256=oBhEQ83SbIlYc7QSG6Cqk48QqdYSlxcSMWbc-2KXlXM,1191
@@ -122,16 +123,16 @@ sglang/srt/distributed/device_communicators/shm_broadcast.py,sha256=IrSrnpZnii0E
 sglang/srt/distributed/device_communicators/xpu_communicator.py,sha256=ajW6132BvA6jkeipEIgN27TFycI0U06Ih2Z8WNjlA4s,1593
 sglang/srt/entrypoints/EngineBase.py,sha256=yKN76witT2jz1zhmLHmPNLGMpK2UiOTaKQ2KPD8l99U,2594
 sglang/srt/entrypoints/context.py,sha256=aD-94xkD0komuGO5gtYUoJKCHdc4hAipMxQt04yVRGA,8030
-sglang/srt/entrypoints/engine.py,sha256=2x5OhR3CEOrPwMCKIkRxXk0LLGZI4e8vamjGkHQ5g3s,33151
+sglang/srt/entrypoints/engine.py,sha256=Dt84ZOeuNKqjlC_PNU6dt6Cr4pFJxXwHdkYEjInmaz4,33151
 sglang/srt/entrypoints/harmony_utils.py,sha256=01T-A5GBUm2b306PcxNEg2rfx4cykBcqNYrzcXTWBlc,13590
 sglang/srt/entrypoints/http_server.py,sha256=B_nV6dqQWOHj2t_r0_rip30AMfJ8ZTgnIUJLeVt3qQw,44249
 sglang/srt/entrypoints/http_server_engine.py,sha256=_--j4U04OeJLlnnv1f0XmCd_Ry0z1FlhkrbePX8rYV0,4938
-sglang/srt/entrypoints/tool.py,sha256=6fyXArrLqDsZO3rTtPfZ_6ngYLfTMRyyTSvESqos5tA,2678
+sglang/srt/entrypoints/tool.py,sha256=aAXEjBqkr5j6_IgRL8fZCKgoDPY2TikM20-LQXU9TaI,2709
 sglang/srt/entrypoints/openai/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sglang/srt/entrypoints/openai/protocol.py,sha256=S_uMVVg8LwxN__O6corf60VqayNQMAGM_BxxU916RGE,26499
+sglang/srt/entrypoints/openai/protocol.py,sha256=Tm2dVpIHozcxXVyQuPxI6WTofdz0UfOpA9VeIvMExRU,26740
 sglang/srt/entrypoints/openai/serving_base.py,sha256=epD6J7K72p6TP8o2dGYf-a76nXeF80vFhPRpWts6684,5445
-sglang/srt/entrypoints/openai/serving_chat.py,sha256=uI9NjmB-Rqnvsm70i6D0zZaJnacxkOuMYnxuBWMKPdc,41008
-sglang/srt/entrypoints/openai/serving_completions.py,sha256=ByM6SJEDA-y0VV2MdkuLMdlb4mDRtD81tQtyMfB0SdE,17088
+sglang/srt/entrypoints/openai/serving_chat.py,sha256=t5rZ0LxB1VXeclKex3cbzik4zmJ4ke1XyDmZOtRdJIk,43120
+sglang/srt/entrypoints/openai/serving_completions.py,sha256=7aamrEWd7hu57c9mJ86Oe8AeI9qHF8TM74T6Gy-Efzw,17844
 sglang/srt/entrypoints/openai/serving_embedding.py,sha256=hqPt2ELpT1yQ8sBXo801aLJ3sExoPXs_K5ZQtC2vpAs,6274
 sglang/srt/entrypoints/openai/serving_rerank.py,sha256=8n6y5kC6UhRpWrDtzH0eIp0dznW12FT60MLCR4bvD2U,3333
 sglang/srt/entrypoints/openai/serving_responses.py,sha256=aTdC98ZvqAjmGn6VntVw_c8OwtVYZnsewDjQGtKZ8q0,53398
@@ -152,11 +153,12 @@ sglang/srt/eplb/eplb_simulator/__init__.py,sha256=HIZaXqsvsOCMMJ81dSs1afSGZ5G6Oi
 sglang/srt/eplb/eplb_simulator/reader.py,sha256=OlsWqsuXqVQY7PfvvrP-vsl2Ww2Kg4vD_K9BltM_QHo,1828
 sglang/srt/function_call/base_format_detector.py,sha256=FKBzsGFHfNl3Wvccsdvv9ftjbCZIyyjOLj2kHgKvX9g,15901
 sglang/srt/function_call/core_types.py,sha256=wLUsW8mUA-i-ISz5QUDL7Ejq72W-K1HIhFa9Wm-_oA8,786
+sglang/srt/function_call/deepseekv31_detector.py,sha256=hpRjjv219mH0v1PQpMvj_gezrP7PwvLiNbBKOTCSC3U,9731
 sglang/srt/function_call/deepseekv3_detector.py,sha256=DhO3tUX1nvFlfRkHJ1Vm_7OMXFFTvWBA9u9SZKFAwuQ,9710
 sglang/srt/function_call/ebnf_composer.py,sha256=f5Lqn3mtTtJ7LCJXAurVw2k4Zp5LgbWywQv7GsxSoPw,15000
-sglang/srt/function_call/function_call_parser.py,sha256=YHt-5qIeFwueOQCiXzfKKwDXXkQ7AaNYvJlmNhMSt34,8211
+sglang/srt/function_call/function_call_parser.py,sha256=55bbToi2wWGmahfyDX1YeFpndNCwTa23ZdNsawMidbc,8333
 sglang/srt/function_call/glm4_moe_detector.py,sha256=FS4Jhyt8_2xTTytNsQhzlNds2VQO_JIp3tKyCf7PtF0,6939
-sglang/srt/function_call/gpt_oss_detector.py,sha256=F9RfZ_NFyb0HkwNOK_B4fk4etsfsj505bvcQUQuPrKc,14201
+sglang/srt/function_call/gpt_oss_detector.py,sha256=uVhqYA5vATUIMaLPQX81qy2Fulwo6IlwCA_PbUg3gZY,8299
 sglang/srt/function_call/kimik2_detector.py,sha256=u6GXTAcvqWstkuDgxZMeKlpp9zu3V4OTZ8yaplLxEGM,10109
 sglang/srt/function_call/llama32_detector.py,sha256=trZja7IXc2IWUko5PRj7p4pRU6XCVjtxr-7qJNr0EWo,3647
 sglang/srt/function_call/mistral_detector.py,sha256=Ts2HrJdu5r2lCCMVKPZfnCds9pt1K5H1EaOQrF_HD60,4686
@@ -167,7 +169,7 @@ sglang/srt/function_call/step3_detector.py,sha256=SAibrispu0peDQzqOG2-V93KGHtV6U
 sglang/srt/function_call/utils.py,sha256=__ImDF2kNyoLWsYO5RYoryvy1mmgEjnjXlCvLv-uLCM,1695
 sglang/srt/layers/activation.py,sha256=uhfhh10N5iLoLEBCvFPRjCfDxu5jXe1ehHJb3ISIjig,7840
 sglang/srt/layers/amx_utils.py,sha256=1mENgHK2B8mgaD1oMtgbZ15Jmy_Uu1QueBmo09Ff2iA,2865
-sglang/srt/layers/communicator.py,sha256=FyumhAAE0qJBk9v7B7ad_dzmSuM3KCk-dEAz1lWBj78,22660
+sglang/srt/layers/communicator.py,sha256=RAP_oBwHBZdc5Knnb9FRozStODWc4toqDjU4wvekzHc,22625
 sglang/srt/layers/dp_attention.py,sha256=35GQIpO8HpF9mtLQSJe2RW2a5rblTLvzAbr1CL0PbSg,15286
 sglang/srt/layers/elementwise.py,sha256=PjYv7Sk-zKbG9WiGlYDKsxvNrnxT2N1Q8bA2BWvA1Uo,18771
 sglang/srt/layers/flashinfer_comm_fusion.py,sha256=hnVVS_VsWukTYhREubWs2KcJ1GjVMEI9ECqxqOMXZv8,6729
@@ -181,17 +183,17 @@ sglang/srt/layers/radix_attention.py,sha256=MpugwaqiY2zPTeg7FVW41OljljCXmYVvHZ2M
 sglang/srt/layers/rotary_embedding.py,sha256=MZ-P1wcVSr2YA0Iz-4qKIyL8G5p_-nVJU2EA5MVISgM,73102
 sglang/srt/layers/sampler.py,sha256=cgtAtsxcllqhlrxk5ktiBEAeeVLZb-5wWb9NlGzkNYg,11199
 sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSahyMfA,4063
-sglang/srt/layers/utils.py,sha256=NEWsf7OlUgcX42WHXxANduDL0ByBm3beirCp8gvHDRs,1437
+sglang/srt/layers/utils.py,sha256=E7PHisGiD9ZmLS4RWqCUZWxAQJbfc0WaTfRZ37gUlPI,1077
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=x0421kNUgnAb5LnkJR6O67vGZ0ikSD-ThluV3QGsZ_8,22673
 sglang/srt/layers/attention/aiter_backend.py,sha256=XfZKrPFL1feerRMRL4jFoBGtbtoOD9gzKFCx8BomfJ4,43146
-sglang/srt/layers/attention/ascend_backend.py,sha256=ySS1XX5_tnBOG9uBrQRzNmWD1fuGkjJjnwtOjimXv6A,12779
+sglang/srt/layers/attention/ascend_backend.py,sha256=3I6ZCvDMs0QO9ahcg6egt1xLHVD0oLKtcyBG1mU1ciI,18158
 sglang/srt/layers/attention/base_attn_backend.py,sha256=_vM05ddWy6SaoqWYPR8w5lflSE4MhcVFR271Q9EKK8E,3600
 sglang/srt/layers/attention/cutlass_mla_backend.py,sha256=PB5BjbJHhmTg_t0kXfarTI6C3pPvKhBAqvCSGGajou0,9846
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=2ZRL_gYz14idoVqQzeQ6N77nXer0f_8_TUYw40XUUz0,9161
 sglang/srt/layers/attention/dual_chunk_flashattention_backend.py,sha256=hCATQEhBATZpPhBMF-cCM9r6nxXA0dr3jOT_hTFS_T0,68656
-sglang/srt/layers/attention/flashattention_backend.py,sha256=bB4lW2TqTr-j0TOk1nxOnCytRVtAksRECxqIRKx-VKQ,94851
-sglang/srt/layers/attention/flashinfer_backend.py,sha256=J5Q5PeMyMo0NZa2BLgVWJZHST4UQCrifP6Rrv7iXlXc,54215
-sglang/srt/layers/attention/flashinfer_mla_backend.py,sha256=5dFkb7oD62MITS0V4Bc8b-NfFIJR3Lx3EroDTaiLm-Y,40985
+sglang/srt/layers/attention/flashattention_backend.py,sha256=3Jgapx8tVh2PbPVi6e5TX6eBttdOSm2V53TDbZRixr8,103779
+sglang/srt/layers/attention/flashinfer_backend.py,sha256=KLN2t7qn0UtRMeNO9OPLgwyhj23KPAwx_z_OmJSN4uQ,54197
+sglang/srt/layers/attention/flashinfer_mla_backend.py,sha256=GNHvib_PaUO-JFmUcwRADzF7-4v61KFbvk0yEExMdhk,39556
 sglang/srt/layers/attention/flashmla_backend.py,sha256=5iSic5ho-lkXNas9mR3uLbXbEl-do31gc8gjR7-a79k,20711
 sglang/srt/layers/attention/hybrid_attn_backend.py,sha256=dwbxscTGMOnwfxxPpmGdbSiQ0kQRHnH0pdpKHA-_vqc,3181
 sglang/srt/layers/attention/intel_amx_backend.py,sha256=hH9lxXs_o66_IVs58oPhRCOXGUrHbXSz82FKtHr9GO8,4092
@@ -201,7 +203,7 @@ sglang/srt/layers/attention/torch_native_backend.py,sha256=RsI4lYiQJAj-Xy0vKkzby
 sglang/srt/layers/attention/triton_backend.py,sha256=fvcJLaJaM8JE9Q9Nkmn0dzdYnnYlwELCoxQGMoqS9Gk,41858
 sglang/srt/layers/attention/trtllm_mha_backend.py,sha256=8MxCAzqsNuvO3mhFqTMXCJ0Tt0_82kFbBHie6m0ieDA,27588
 sglang/srt/layers/attention/trtllm_mla_backend.py,sha256=pkXARtA9ISC3RL82mqjY7vzfsK-u1rvMRBcOn2n0J_o,19101
-sglang/srt/layers/attention/utils.py,sha256=lIlHKwSYcS6R3RVWuoLj-5RnJAMMtmX9LxLAE-IgZKs,7025
+sglang/srt/layers/attention/utils.py,sha256=lx_jNw2VwlBe2VnDb1eXRRdN9WArw9rdWHTHqndUEFk,3174
 sglang/srt/layers/attention/vision.py,sha256=PZUJuswPWvJ34NTtxJkcXnusojDv0pO0UhrsjL4SWYU,20268
 sglang/srt/layers/attention/vision_utils.py,sha256=6hPPAt8p3SOm7tgHNUesnMLxvQiKbaOM5EK6K-ig5dI,2771
 sglang/srt/layers/attention/wave_backend.py,sha256=_rjyALSl5huNtIGSe6OJlBrC04LhwBBu62bgjlkavkw,23010
@@ -215,17 +217,17 @@ sglang/srt/layers/attention/wave_ops/decode_attention.py,sha256=C-W1hvswZ7ggDKp4
 sglang/srt/layers/attention/wave_ops/extend_attention.py,sha256=aPNVsrhJD3OW7YF3jRm59IIawgA-bv_6GT91UPf5x5Y,3957
 sglang/srt/layers/attention/wave_ops/prefill_attention.py,sha256=viTUit0rxjVV5UahuwmVAhmL66YRFJD4g1YhH_oRBu8,2344
 sglang/srt/layers/moe/__init__.py,sha256=63TxUpSiUpVg1SDY1zdlTg3WFJzAc7WSndOViOmUv4E,835
-sglang/srt/layers/moe/cutlass_moe.py,sha256=8WWLlEQF5pQHXGsFmmyLCNBo4i1kfI8a_CKg8ruztMQ,14800
+sglang/srt/layers/moe/cutlass_moe.py,sha256=JKJED-4709ndP5AwhQ7Vi04GJjw5d9Xl_mWOsZPZ3U4,14298
 sglang/srt/layers/moe/cutlass_moe_params.py,sha256=9NRCmgP_Ug3gGqCcpi-x-QRbLjCNpw8792gKXwZsbEU,6522
 sglang/srt/layers/moe/cutlass_w4a8_moe.py,sha256=rWyHMg0kYZhIdLWTJeC7pqRG9ywfeKqi47-OekNeF4Y,7306
 sglang/srt/layers/moe/fused_moe_native.py,sha256=8SAToE4B-22H5JsENZgJ1Io6QfE5-D9ItWLtbGksFQQ,3372
 sglang/srt/layers/moe/rocm_moe_utils.py,sha256=07Z99bTV3B-b2Cbm-odhGpx2twxtnVpYvaDMBE1K3LM,4555
 sglang/srt/layers/moe/router.py,sha256=eUNu_Uz5VB2FOZzZyYuZo5pokCVBS17_fcjHQbmvDSE,12181
-sglang/srt/layers/moe/topk.py,sha256=rD7bO7--LPT6NvQXksBwG8tAkCqJoaxzF5L03mvbXf0,30378
+sglang/srt/layers/moe/topk.py,sha256=wxCkMzm4FdCQM5-Gr5eS0qTTbKxntU6dFKU4Qn9DQYc,30377
 sglang/srt/layers/moe/utils.py,sha256=omh9E6sF-KtrELEsE9y01Ash7FfQlstyE-97UtpH8qk,5998
 sglang/srt/layers/moe/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/srt/layers/moe/ep_moe/kernels.py,sha256=7MieYfBctQjqfBoKun_QUyfloIrrgt5tjzmMkMoY5Ys,44107
-sglang/srt/layers/moe/ep_moe/layer.py,sha256=tJwbzSKT0qMRF2E4BDwyzFhUCBd1-KxInJxqmgrP-y8,28191
+sglang/srt/layers/moe/ep_moe/layer.py,sha256=Ro4SQ9peGw4-1oHFHsw8HvFb2yEdTkB8JH7MnSV1hoM,27816
 sglang/srt/layers/moe/fused_moe_triton/__init__.py,sha256=4NXZHbCw-G-uSnNUj4up0yh3xBDPnT-x0pdoIr0lku8,831
 sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=JKp89Cdl8a85tKzmJa4ah5VtxZzRNBSbY4cnsAtnvio,60803
 sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=jub7moptP9UjyRYJ2gPvY8HEnE41d_oFbZgjVvNsJxQ,39409
@@ -418,6 +420,8 @@ sglang/srt/layers/moe/fused_moe_triton/triton_kernels_moe.py,sha256=Ai06BZ7uxMnk
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=352,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json",sha256=RMYwqtB7hYeaCyuZ1feVnRhBGlN4jMcmNQX2hqUqBlk,3254
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=129,N=704,device_name=NVIDIA_B200,dtype=fp8_w8a8.json",sha256=W8C1GtP4K43SK9128U52DD5WWofvPleAJE4us2Qju1k,3251
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=161,N=384,device_name=NVIDIA_RTX_PRO_6000_Blackwell_Max-Q_Workstation_Edition,dtype=fp8_w8a8.json",sha256=RMYwqtB7hYeaCyuZ1feVnRhBGlN4jMcmNQX2hqUqBlk,3254
+"sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=T6kCtCUytCxiW8JOLfEF3YgdCD9_mKccJ85IqT0b8LU,3253
+"sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=257,N=64,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=QEk8bE-GieZltTiv3ig__ScM9Q_4K-c75UzIuDrGmPA,3232
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_4_0/E=384,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FEnjuvvky-cvW1iq8TPrfGOOH8y4rOEx4jlYZ4tqbDs,3253
 sglang/srt/layers/moe/moe_runner/__init__.py,sha256=piJQmSz7BOBaxVd0gGMgRLzaD21r7PH1NEDOK5hktYk,97
 sglang/srt/layers/moe/moe_runner/base.py,sha256=L7ZNcq6fQxMXQZypcxWKIe1j9ajoOKCsFRrF-UlHVaA,364
@@ -430,9 +434,9 @@ sglang/srt/layers/quantization/awq.py,sha256=-W4c4c5Df0h7d7mQKQFSJ9ZAxCtyWiOxpu5
 sglang/srt/layers/quantization/awq_triton.py,sha256=4XwJbRNlFmlvjbMGtWF1Wvk26Kf96162vNEmuI5Kars,11737
 sglang/srt/layers/quantization/base_config.py,sha256=taPmvTT2ZMzXShSHoAAYEAjF6_bvIcUY9Oa_ED_SpkY,7066
 sglang/srt/layers/quantization/blockwise_int8.py,sha256=juOCfNFreRrSNwT3IWwj1f7AzlGDsR7WVl_DOCUznvU,13584
-sglang/srt/layers/quantization/fp8.py,sha256=o_tzYZfgspAuoimcrr_XqcJ9lgBZDuONI8baHtvn3R8,49661
-sglang/srt/layers/quantization/fp8_kernel.py,sha256=q1dTlGEtdQB3mSvNC5x2lvN23N4WF5eVFCrEf4dQjlI,54875
-sglang/srt/layers/quantization/fp8_utils.py,sha256=a4sJaNDCRhS3y9maByEmXj0j3h1hklGw6jiPqEpj_-M,27971
+sglang/srt/layers/quantization/fp8.py,sha256=sWvsA1Ch_4JCTgZeM-ncf0jJUTQ4LF9_ruJNZOuKc9s,49634
+sglang/srt/layers/quantization/fp8_kernel.py,sha256=Mgigb6GmF66oCrgjS97IvDDSacNYb0u-VYAf4FChhqM,54849
+sglang/srt/layers/quantization/fp8_utils.py,sha256=emYXdBdIi5peGtapIREr59IteiQcNTAgWULkU7Npzzo,27969
 sglang/srt/layers/quantization/fpgemm_fp8.py,sha256=dq7rJz-QB9LovmZZLBMK8YJEOr05wjVrtWLmSTkXjYw,7004
 sglang/srt/layers/quantization/gptq.py,sha256=fBcC-9v9GhLvU_oquKvmjD6O_p725SZPPOUtmDPz66E,39118
 sglang/srt/layers/quantization/int8_kernel.py,sha256=GfRC9FOn9exNvK4QHbUeBj3Hhv32VcyGphapFPt5b84,12625
@@ -440,10 +444,10 @@ sglang/srt/layers/quantization/int8_utils.py,sha256=YK9CS-lb_n91kNCTKK5o5apYF31V
 sglang/srt/layers/quantization/kv_cache.py,sha256=_9pF5rwvB7ta6Gdc5YKVVGbNzYwqmhIx4TrX1-xnodQ,3261
 sglang/srt/layers/quantization/marlin_utils.py,sha256=aM3GU7b7rHNZE-tZjRSrtOaPWzSaMcLxck4he8MlEVI,26344
 sglang/srt/layers/quantization/marlin_utils_fp8.py,sha256=iSjq-ikPOTU5plrWNTW4MletKLySIrNl5NhQmY6FyAo,12508
-sglang/srt/layers/quantization/modelopt_quant.py,sha256=1d7ZGQ4_lsfQAVzkNPN_DeVSwWWWEPzWXywO8VIvw-M,52617
+sglang/srt/layers/quantization/modelopt_quant.py,sha256=k0F6b-bG5wv3zL7kipqCrQHEZ8QjfIDThnU9Y3mFNTA,52566
 sglang/srt/layers/quantization/moe_wna16.py,sha256=brU0nG6StgjGuuhOjpeoh-wJVZ1SUgbuHhiDR1hibbw,18905
-sglang/srt/layers/quantization/mxfp4.py,sha256=v6GlGpvJT2GH75djiJHmz4r-E3SKhvpVD_-UJrHKfUM,31410
-sglang/srt/layers/quantization/mxfp4_tensor.py,sha256=I1hh_uZqIqkgFnNyjaP17cDMLMV5rs3aFytg91MPhzk,5357
+sglang/srt/layers/quantization/mxfp4.py,sha256=xCTaEtL9KLMzHjMP4snItR4sXKACTk1TCNQryRzmaXE,31195
+sglang/srt/layers/quantization/mxfp4_tensor.py,sha256=bGeCXlIOhmFZFGWp_lITkIRDEgGUaLpawDvMy1cTOmA,5389
 sglang/srt/layers/quantization/petit.py,sha256=74Jn0eICq23v_b4rEctDaYTINtbqkdM6IXycT-_FbbI,8954
 sglang/srt/layers/quantization/petit_utils.py,sha256=-gy4zMhqNoIA1R0n7-5C0efV54jHonCUgPDUUFjGsyM,3245
 sglang/srt/layers/quantization/qoq.py,sha256=jM96uJfKz3vo1B6SSIrQXNT3vnD3UJrepK2eY-tSQU4,8139
@@ -613,9 +617,9 @@ sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a
 "sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_H20,dtype=int8_w8a8,block_shape=[128, 128].json",sha256=FImA-TJ_tQDjqwoNWxS--sRDoKDXf9gamlME3tkxH58,3252
 "sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FFBjSWlpKXMxfAUUYUqXbOK_Hd7qBeBsfbcaa9uB4qY,3249
 sglang/srt/layers/quantization/deep_gemm_wrapper/__init__.py,sha256=esJMd0Yuj68t6QYOpmIFuiWP2J2dxTMC4bRBNH0Xk6I,26
-sglang/srt/layers/quantization/deep_gemm_wrapper/compile_utils.py,sha256=FaHKgY9y1oi3zlCmkorSK4r4dhPy7NiviVIhtabvzjM,11536
-sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py,sha256=LPO728y32uHQOGfk8U26jDIbGlAcQdD-ZYaqfboyhzM,853
-sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py,sha256=On_uvP1ThxgQFAzOWKoteM0vIOHqmtI127A4qbPtmbc,3264
+sglang/srt/layers/quantization/deep_gemm_wrapper/compile_utils.py,sha256=x5CcDGvfvJStt2yS1tlfOup7UcG_0L7iHPMF33mR9-U,7908
+sglang/srt/layers/quantization/deep_gemm_wrapper/configurer.py,sha256=WAMl4vrG6fXw2ar-svT-UO-mgUzwGV6dumkUiUxuTaY,876
+sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py,sha256=vCj5vdAshEB9mAgSUYXhgJ0bd1Ithmu_n4-m_IWUbd4,2531
 sglang/srt/layers/quantization/quark/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/srt/layers/quantization/quark/quark.py,sha256=EqQvWzAc6nd_yl6UIMIwmozKif4I8fRNfBum040UUEQ,14613
 sglang/srt/layers/quantization/quark/quark_moe.py,sha256=yLRmytEwBMZd7ShzHVX6-w_8koyiRrvIxKx8g02qRZw,6789
@@ -626,7 +630,7 @@ sglang/srt/layers/quantization/quark/schemes/quark_w4a4_mxfp4.py,sha256=gMKmrpim
 sglang/srt/lora/layers.py,sha256=3RwD5TC4zO8A2Haz7ScIT5vc_neU9RHM3DoamKttVkw,10838
 sglang/srt/lora/lora.py,sha256=lPPEJavrfANXDzdmKEg4hbgmKIiEr3NU7KNJeaaKTB8,7393
 sglang/srt/lora/lora_config.py,sha256=qDgMTx_69jyJUl29O5FxLzYa0BMhqYVXWXfyyVOvGm0,1684
-sglang/srt/lora/lora_manager.py,sha256=FPb-S0qDpCEpO3tv8fuqI7ULznq_UaC8Db2ePJljEBc,20765
+sglang/srt/lora/lora_manager.py,sha256=HSA-ZnL2NUytos4DxzawrZ-1yoaoL42Stn5Y92lXf7w,21781
 sglang/srt/lora/lora_registry.py,sha256=osbau-sQZ7BuCeGGpBMFy4w40GO5FGqRgRwDyxxgNmg,8167
 sglang/srt/lora/mem_pool.py,sha256=FDf9qNEWrZ3yTtrUEvbiUN9FUawRZBlK9l9A2ou5dkU,11242
 sglang/srt/lora/utils.py,sha256=DePsHi-0S-eCerQeWXTW9U7sG2397w_Bhfs0TAsu4xk,3904
@@ -637,25 +641,25 @@ sglang/srt/lora/triton_ops/gate_up_lora_b.py,sha256=QURCYxHNR8Ls4SQtt3dvdgjvdDVh
 sglang/srt/lora/triton_ops/qkv_lora_b.py,sha256=z9BB84Xu_MjwCu5v9Yvji-FkIzfEjDySAFybc4WYv7A,6483
 sglang/srt/lora/triton_ops/sgemm_lora_a.py,sha256=Ksova04wgeGsFqGOXWqJtMYaHgyUYcx8VU42BZQOkVA,5129
 sglang/srt/lora/triton_ops/sgemm_lora_b.py,sha256=VqCAFvUtq_l-0RGIkx3W_fzD55QcW2FAcqpvSHOBFp4,5115
-sglang/srt/managers/cache_controller.py,sha256=EJX39oikAM5YOkTpqoF-Rg-dFnBW-eWuiFrE88q4_7k,33341
+sglang/srt/managers/cache_controller.py,sha256=iSGfeS3IATQ6bcYEoDVZZvybP7KAJZGTX_SXw46JalU,34560
 sglang/srt/managers/configure_logging.py,sha256=8sNXZ2z9pBWOwn-X3wyz013Ob8Nbm1zDxRkxoZjH-l4,1633
 sglang/srt/managers/data_parallel_controller.py,sha256=VvEkidmYCGAFTv6upEHf_tyTvkM8xLhCyOnLJGWL6BQ,14592
-sglang/srt/managers/detokenizer_manager.py,sha256=DaVDYaKYLcaTU3Ll4ZJtOdWQGpwbgYcba21JOWOlUBU,10961
-sglang/srt/managers/io_struct.py,sha256=S7MGa5OvhOB2odBI8eC34fqzMX1QrhNoBPAF4rkGFLc,39403
+sglang/srt/managers/detokenizer_manager.py,sha256=I1V4PqQo23stteiwyJXQryP2pAauNNq-pbeffhTDgQI,11244
+sglang/srt/managers/io_struct.py,sha256=xXMgXavOnQAPA3vI801nzZ9ha6QYMNmai5uSWcLQ-zU,40023
 sglang/srt/managers/mm_utils.py,sha256=0Kr3VJKINr9o4k2Tg2KN0Zb9kxRYBd4dxd0ImTbKHw4,29351
 sglang/srt/managers/multimodal_processor.py,sha256=cnWpu2G79v1a6FJB_FriLxESgGUbfC3GptLmeRVVgew,1801
 sglang/srt/managers/schedule_batch.py,sha256=9lhBnf-siQjsThD8FRyhiF50N-LOE7dfoPr1uzklnjQ,77170
 sglang/srt/managers/schedule_policy.py,sha256=vh9BQW9tBv80LW4JApLE6smU2m4gy6cAEI25HVXiS60,22383
-sglang/srt/managers/scheduler.py,sha256=RXVUuOLyboWV9NR9U12EhfcJeKfW34fVLFmZGd4OFNA,107583
+sglang/srt/managers/scheduler.py,sha256=RQuW5MrfSkl1F7TbioVuIeYujiRp_leANRhzxyRhTHU,109467
 sglang/srt/managers/scheduler_input_blocker.py,sha256=zP8xU_UmU2H0AB6sEqvivDcDB1QDgTauNDYDIJ7Nez4,3683
-sglang/srt/managers/scheduler_metrics_mixin.py,sha256=-1IOE30eOoTvvpz-Xg9J4UcamwAJle2-odFJ9vrinLE,9341
+sglang/srt/managers/scheduler_metrics_mixin.py,sha256=31BbukSyUUPMNHLqRlS5sEiAv7Gi5VHFQ1TgxvcAdcw,10054
 sglang/srt/managers/scheduler_output_processor_mixin.py,sha256=e_l5TRm05liINSAeZU-YPRpJXcutmEWYV30F1vPaxbs,31345
 sglang/srt/managers/scheduler_profiler_mixin.py,sha256=A67lC_U7HUCf_znatXK51orC8xeFNGouJ866HH_Y-yc,11402
 sglang/srt/managers/scheduler_recv_skipper.py,sha256=8rVQCBjn9f26Th4XbeUS3fEsvo1GI029Me28wMjcKRY,1103
 sglang/srt/managers/scheduler_update_weights_mixin.py,sha256=R81fU-0aIJoi5URm9IA-n2sEjvn3phGXOhzJALZjiOM,5467
 sglang/srt/managers/session_controller.py,sha256=FGADRGSkRmS76o_BQ2fivpgw0Q5Q-K0Ws3aezH41YSc,6041
 sglang/srt/managers/template_manager.py,sha256=m_cy_KY5bOkDlh6IB9ocisy8N5r7abp4nfFnhruNxFg,11947
-sglang/srt/managers/tokenizer_manager.py,sha256=iy7aH4vugjyHFZl3qAGKWGovuNTn-aokUNb6BShy_1E,84893
+sglang/srt/managers/tokenizer_manager.py,sha256=xW3fNwocCtHPAjc1EEegpzYqpeMfOd-SFvIDfUn0pKw,86156
 sglang/srt/managers/tp_worker.py,sha256=n_DqSD4JqOJYjdZrchOpmpuTazRw-AUrs6RPQYuAPyQ,11933
 sglang/srt/managers/tp_worker_overlap_thread.py,sha256=NXjrBT7lvdUlDDNFzEQyFYCerpqj92sNl2YynGhQj-Y,11118
 sglang/srt/managers/utils.py,sha256=WbUdADEdiSOoQNfwjfU64mNaXoskEveBO1HP_t0FJtQ,5687
@@ -664,31 +668,31 @@ sglang/srt/mem_cache/allocator_ascend.py,sha256=i0KPCcFlzGkCIjW8ph3IPkBXAWTqq4-A
 sglang/srt/mem_cache/base_prefix_cache.py,sha256=hLS2ncTMAz7Kpdk5pNwn5c6g8b61_K9OCBL4Vj1xsYc,2801
 sglang/srt/mem_cache/chunk_cache.py,sha256=xQjfmpijkkNRuqfwYMhoiV2nJAviABX4uOSLE-yyVus,3184
 sglang/srt/mem_cache/flush_cache.py,sha256=GYcxmNXh4hsMpFfNOuCTpKilW7guZwTtAg_usVeM3J0,979
-sglang/srt/mem_cache/hicache_storage.py,sha256=yjXcsoKdjLgqG68E7cHSAQXq08EHgAASr74bZN0Sg7M,6695
-sglang/srt/mem_cache/hiradix_cache.py,sha256=J-k8HYwJuZlKAFxKWSWvDgaAUp-bgANWH8rr46chnCo,28173
+sglang/srt/mem_cache/hicache_storage.py,sha256=hs8jp88UEOJKsNVsqp_z0GbmRrWs5oQSZDGMUGAUDDk,6974
+sglang/srt/mem_cache/hiradix_cache.py,sha256=yhBuVZKqpko0Vd55aJd34Fqk-xeCbo6woD5EMOoeA54,28574
 sglang/srt/mem_cache/lora_radix_cache.py,sha256=FCA1d-TSqXYPrz9-m-cRrxse9RtwALXltGNuzR-PnB8,14354
-sglang/srt/mem_cache/memory_pool.py,sha256=pCcrb7-LNqlVtUonu8X8KMRZZdIgzQfi7ia4QW2eGkY,37086
-sglang/srt/mem_cache/memory_pool_host.py,sha256=to15nKPDPq_PwwcIDESUoCMUEowLHH6UMZQmBN2p_8A,26205
+sglang/srt/mem_cache/memory_pool.py,sha256=md3W5lWoa5-g0XLHyYI_EIqcW6yijN7usnbmbD_C_GY,39396
+sglang/srt/mem_cache/memory_pool_host.py,sha256=cXNn4C0m-MXM6g36uF1Dx5Ld9zP9ZAhlTngohQGpfUc,26215
 sglang/srt/mem_cache/multimodal_cache.py,sha256=zPnQLQhBZ6zsUpCQPSoNkrB9EEvpoDQS4mU7c3sRWjE,2171
-sglang/srt/mem_cache/radix_cache.py,sha256=VdsMPUkaBUjLDTzMqKCwBr1CDP0GlbS_8bJCHrONlk8,19046
+sglang/srt/mem_cache/radix_cache.py,sha256=MdX_gYcrif6HJQTIGTD9sA1zOLIn8920lbqTO96TpgQ,19085
 sglang/srt/mem_cache/radix_cache_cpp.py,sha256=1V-XxYC2jxQYUMJBLrlcSMcqhCBo8TC7wRTTVbOaMkY,9269
 sglang/srt/mem_cache/swa_radix_cache.py,sha256=4qt_gS7GuHiekd6VvKkx54dB2GHJzNA_kj71Np6jTDA,40300
 sglang/srt/mem_cache/cpp_radix_tree/radix_tree.py,sha256=tQZpz-H6HxNAQe9mYqt6aX8mfDZyu_fbJTcCNK-Ns1M,7179
 sglang/srt/mem_cache/storage/hf3fs/client_hf3fs.py,sha256=WV5M9UPPLUMMjU3Mpp6HBqyCQpGBuMyj-N5PwQcNo5k,5178
 sglang/srt/mem_cache/storage/hf3fs/hf3fs_utils.cpp,sha256=dDYPtxmL6OFhEyv_GuTgzkTDpBjcD0EK8kphnn7eaGc,1133
 sglang/srt/mem_cache/storage/hf3fs/mini_3fs_metadata_server.py,sha256=OfQMFwbbQg0pbC7rJktg5wz3YFhNsNOjsXCsAjrh6iU,16738
-sglang/srt/mem_cache/storage/hf3fs/storage_hf3fs.py,sha256=WcJKrs8uQ9STKc30HYwd6oqT0fGn4UYXQipuEbDvXPw,13145
+sglang/srt/mem_cache/storage/hf3fs/storage_hf3fs.py,sha256=MlN9RLhd2W81RDSvaPtGm_AP_QKwO_KWUSKXMi_rnL8,12918
 sglang/srt/mem_cache/storage/hf3fs/test_hf3fs_utils.py,sha256=g2h0Woa2CK39pRTigHCVR_hq095dTaC70lCIAIs9THM,1089
-sglang/srt/mem_cache/storage/mooncake_store/mooncake_store.py,sha256=UIHykWDBpf1Kld7Fej5Kr9KPQOGeOoVGwtLup3JnuVs,9663
+sglang/srt/mem_cache/storage/mooncake_store/mooncake_store.py,sha256=TevMG2cd05-uUxDfW1LDk35pLwOwiTfVDdVyw-4lsZs,9951
 sglang/srt/mem_cache/storage/mooncake_store/unit_test.py,sha256=jMdJgU41VscKS-mn_AVZE8LQbb_3MWQ57i9V3OCWLI4,1007
 sglang/srt/mem_cache/storage/nixl/hicache_nixl.py,sha256=bau9Jsp50lpEiy_osgb-dYzeoUkPWM03x7Q8z2BpKaU,9920
 sglang/srt/mem_cache/storage/nixl/nixl_utils.py,sha256=KXkwMrbWwIOUOASINnUo4HyheiMd-93smmwGLv0CacU,7463
 sglang/srt/mem_cache/storage/nixl/test_hicache_nixl_storage.py,sha256=vENVvGAd3PqcQEFA812wPoSBYhRemLr8Sfh1LQa0INU,9609
-sglang/srt/metrics/collector.py,sha256=FH6Wil2lMK7d7q74CfxfdULCaHKDXeAoo8UBZqqdepM,20289
+sglang/srt/metrics/collector.py,sha256=IfFr4RaamT3TPj9DjuEnvkdiBGKcIUCQSIIwd9GeZe0,20295
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
-sglang/srt/model_executor/cuda_graph_runner.py,sha256=vFIR-DMVfyBPQIlHX3spaoPsXbx2RCC3h77vHmhb2e0,33896
+sglang/srt/model_executor/cuda_graph_runner.py,sha256=fZT_CQwhfu7kdnliKu5lYM2tunU0u3wBoIlCxm4FXCA,33920
 sglang/srt/model_executor/forward_batch_info.py,sha256=2KBu9SChR-YwSNbw1mO3WR0P_EGQjFUC5rQfBx6js90,40707
-sglang/srt/model_executor/model_runner.py,sha256=FJYnyFsy6UqX3ZTZQZz0alyVGj8zs9IlSmQn9AfNAj0,78564
+sglang/srt/model_executor/model_runner.py,sha256=VlSgaKfGa7E14wF18xOKte-PPHzgxVczPVzEQ646nwQ,78533
 sglang/srt/model_executor/npu_graph_runner.py,sha256=isp2Plkf4b_gdjptYxJhTJk7JL21Jg-Md3hAYIuSVxM,3424
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=GPKjMkulnIqlX7gdSSpSselsXLpbEtCLscleC85-1jM,60879
@@ -705,7 +709,7 @@ sglang/srt/models/dbrx.py,sha256=117gwrB_o8VlACwDH5YlXkPfuS7t2Mh-nyJvpNpIYxs,159
 sglang/srt/models/deepseek.py,sha256=M7nyHcAbTIs8vc1g6u00oy22-pp704Xp4T3clBzu4xM,17460
 sglang/srt/models/deepseek_janus_pro.py,sha256=td8xGs6ARfJ8AQCYwUhMOZoWigrAs7m3trF5-kXCqik,70418
 sglang/srt/models/deepseek_nextn.py,sha256=uoiJxAVl9DX7bQcjyuouAZz6W-cB1HMubcirTBghyR4,6115
-sglang/srt/models/deepseek_v2.py,sha256=_4TH56N-VT1uguhGHI3DKHm9C9PPDQNaQInER3AhKC0,106616
+sglang/srt/models/deepseek_v2.py,sha256=hgz0N6iX1oIRo0iR5apPA6nVMAmtlQLJHh3OOLR9Vuw,106961
 sglang/srt/models/deepseek_vl2.py,sha256=7X3gI4DE7guUtNJvMLf06jO8MUHKa2Aan8evZg5hsXc,13061
 sglang/srt/models/ernie4.py,sha256=7dP7_d0i86tv8Mp21Ew9RdsFr6TskIkm8UtUd8zFo00,16067
 sglang/srt/models/ernie4_eagle.py,sha256=E7f-ygCcY8pIoyTMKFyVC2uRcG59-_RaQTO4gd5_uWI,7223
@@ -725,7 +729,7 @@ sglang/srt/models/glm4v.py,sha256=WtDvH4691H6IKrcGBrkBpweSm_iFK61bMEfvqpbblmc,23
 sglang/srt/models/glm4v_moe.py,sha256=qgJtWyRI0dy9qcTMXHxhBZp-ghfghV3PLhpOXgSAW60,17313
 sglang/srt/models/gpt2.py,sha256=kclhxEs8oJk1KCyhmAqo7rZqecVGGHYkc-a1WZi3aIk,9841
 sglang/srt/models/gpt_bigcode.py,sha256=1D6bi8Zu760gCRZkvdLHFcg8kCkY35ARwQYaMDtYhl4,10307
-sglang/srt/models/gpt_oss.py,sha256=qXa6BAm8gBTuCSX0ax3Iz4SkIIko0s1u8sVu8kJVzkY,44930
+sglang/srt/models/gpt_oss.py,sha256=bSi0135zmMDtmKDcMiq1UKHa7hGaUfmx3R_ky8jkuto,44934
 sglang/srt/models/granite.py,sha256=8q92shxVPAp_cJDohJATffSGd7Z0Oi-vF5jpY6DlK4s,19840
 sglang/srt/models/granitemoe.py,sha256=j1rgZ62CbBioECjUblDCw_NneDQgY_QJODsI0fqXVO8,13779
 sglang/srt/models/grok.py,sha256=8KCR13LtdnhswJrIwgb0sdFW0OxCA8GQ0fbN8gb856E,40518
@@ -771,7 +775,7 @@ sglang/srt/models/phimoe.py,sha256=EGaKFi02MZP20qFrILA4o2SJ0wb1HoJagFTSqVUbJcQ,1
 sglang/srt/models/pixtral.py,sha256=ZkZrV0mEfmbBdgRko2AvsMFZbAVskgp_5a5p1xCSRN0,16835
 sglang/srt/models/qwen.py,sha256=j0OU3JjznWi_TbayI09yYGM-Obov6mQzJO82Hs0qBGw,11868
 sglang/srt/models/qwen2.py,sha256=TuKblSvilyVoq9en9e9YpwzLGkpDeOA6IWJELOYSzZA,22733
-sglang/srt/models/qwen2_5_vl.py,sha256=3nRfXL-L230Ml_O3tYv5CbU6paT9ZS1QdfdVWblowIk,22892
+sglang/srt/models/qwen2_5_vl.py,sha256=1FpowSDKjuxvPYppOybnMAAh8mgrWJTPJ8d3qjkbriA,22913
 sglang/srt/models/qwen2_audio.py,sha256=SOOsspU7J6eM-iSkms5op0nWhoxJd6QxmwXU2aQBEEQ,7776
 sglang/srt/models/qwen2_classification.py,sha256=dGrMm4ebd30_lBhHOhaV57ig2iOTx3nqB4GEzsrRIM8,2747
 sglang/srt/models/qwen2_eagle.py,sha256=6spFg6JYQGeUXLv2Q5eMkVzZvnqGyvW6SdVBtFMxlTM,4806
@@ -826,7 +830,7 @@ sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py,sha256=QuPI0S85Ki
 sglang/srt/speculative/eagle_utils.py,sha256=H-2aqm3REztUXLnmhi31ck3TWPPMVKc2CUkWBPsFK44,47785
 sglang/srt/speculative/eagle_worker.py,sha256=tvRV7JNvIeOFb3CTF756kRxQm8SZwfA7rCmm39feMiA,41451
 sglang/srt/speculative/spec_info.py,sha256=rhaKG0TzyF9XZYHEWp1jccwTBohSNsUDvxHFtAoOl18,709
-sglang/srt/tokenizer/tiktoken_tokenizer.py,sha256=su5woRjUNJEOGmi41LiUw__LmzyNIC-UiBikk-B8W90,6158
+sglang/srt/tokenizer/tiktoken_tokenizer.py,sha256=KJwr6Xlc7AdHegMTnAVKJlPpW59Z7AE27GTzFL78GwI,6222
 sglang/srt/weight_sync/tensor_bucket.py,sha256=vp_LRrhULNnxiw9Tf56FcEckgUg07lifgv1lasG6y7o,3695
 sglang/srt/weight_sync/utils.py,sha256=43zl3u8Svzj49p7Sk78xp7SWW83mlr3EH-OZFGUGy5w,4157
 sglang/test/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
@@ -847,7 +851,7 @@ sglang/test/test_block_fp8.py,sha256=mOM-eX9LisotPlMLSCvH4XwWMbvNpVOBtt7euXuVTDw
 sglang/test/test_block_fp8_deep_gemm_blackwell.py,sha256=Hnhq4kkyINHb4ONedkp5Kf7Xx0MjOoOBYL8J7ETvOVA,8180
 sglang/test/test_block_fp8_ep.py,sha256=n4X6ZKwuUUbV5Ofjg64ptlaFGI1LbRXDfFiJW1ELHgY,10546
 sglang/test/test_custom_ops.py,sha256=PenQ8zM1wj5xwiVEPVzD37pO-x90aOfFMpCRZenaKsY,5709
-sglang/test/test_cutlass_moe.py,sha256=fhAZlJmTMwhw7FYgSoYggAvXArE8mqluU6qO5NDLHDE,9835
+sglang/test/test_cutlass_moe.py,sha256=ax-IYPm5tZjkZft0q8swHnzerOI4LB4JOLFaKGJVE-k,9629
 sglang/test/test_cutlass_w4a8_moe.py,sha256=E0ffQOR84bEFz6KFNbczH2LRhQqkeC9HPp76u0vzdwo,8809
 sglang/test/test_deepep_utils.py,sha256=749ysTBGNzh6rYUCJhhZBtZpeD15eWTeNHYCytcvZtc,7448
 sglang/test/test_dynamic_grad_mode.py,sha256=L76yUCuk_ymNpXD2CmO8r2GiGjIvD_gtTsuFDs2NolI,1638
@@ -862,8 +866,8 @@ sglang/test/attention/test_flashattn_backend.py,sha256=_rTG849FwQdVTyGKkqhczaOqn
 sglang/test/attention/test_flashattn_mla_backend.py,sha256=g4O50WblTpM7_Gq2b76k0i25_z01BOUBQ4i6PmyxpO4,10774
 sglang/test/attention/test_prefix_chunk_info.py,sha256=hpoDe2wfSa6RlUbfyri_c0iyBTb35UXGL9I2Xh6jamM,7772
 sglang/test/attention/test_trtllm_mla_backend.py,sha256=ilrX20-1hVo0nmKN0GTqzKoUmDarTXGS8nWJ6c8RHVY,41792
-sglang-0.5.1.post1.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.5.1.post1.dist-info/METADATA,sha256=IgpodN0t3cCFIUqlTb5l0iNjc0obANmUVH_Wv3j5i8M,27803
-sglang-0.5.1.post1.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-sglang-0.5.1.post1.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.5.1.post1.dist-info/RECORD,,
+sglang-0.5.1.post3.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.5.1.post3.dist-info/METADATA,sha256=5F018jFs9QId3jVlqacEYemDaNTGhkFtp_80TEVv_6I,28168
+sglang-0.5.1.post3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+sglang-0.5.1.post3.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.5.1.post3.dist-info/RECORD,,