sglang 0.4.9.post3__py3-none-any.whl → 0.4.9.post4__py3-none-any.whl

sglang/srt/two_batch_overlap.py

@@ -341,15 +341,18 @@ class TboDPAttentionPreparer:
 
     @staticmethod
     def _compute_global_forward_mode(forward_modes):
-        converted_forward_modes = [
-            ForwardMode.DECODE.value if x == ForwardMode.IDLE.value else x
-            for x in forward_modes
+        forward_modes_excluding_idle = [
+            x for x in forward_modes if x != ForwardMode.IDLE.value
         ]
+
+        if not forward_modes_excluding_idle:
+            return ForwardMode.IDLE, False
+
         forward_mode_agree = TboDPAttentionPreparer._is_all_same(
-            converted_forward_modes
+            forward_modes_excluding_idle
         )
         global_forward_mode = (
-            ForwardMode(converted_forward_modes[0]) if forward_mode_agree else None
+            ForwardMode(forward_modes_excluding_idle[0]) if forward_mode_agree else None
         )
         return global_forward_mode, forward_mode_agree
 
@@ -542,6 +545,7 @@ class TboForwardBatchPreparer:
                 tbo_children=None,
                 global_num_tokens_gpu=None,
                 global_num_tokens_cpu=None,
+                dp_padding_mode=None,
                 gathered_buffer=gathered_buffer,
                 global_num_tokens_for_logprob_gpu=None,
                 global_num_tokens_for_logprob_cpu=None,

sglang/srt/utils.py

@@ -744,9 +744,13 @@ def load_image(
         image = Image.open(BytesIO(image_file))
     elif image_file.startswith("http://") or image_file.startswith("https://"):
         timeout = int(os.getenv("REQUEST_TIMEOUT", "3"))
-        response = requests.get(image_file, stream=True, timeout=timeout).raw
-        image = Image.open(response)
-        response.close()
+        response = requests.get(image_file, stream=True, timeout=timeout)
+        try:
+            response.raise_for_status()
+            image = Image.open(response.raw)
+            image.load()  # Force loading to avoid issues after closing the stream
+        finally:
+            response.close()
     elif image_file.lower().endswith(("png", "jpg", "jpeg", "webp", "gif")):
         image = Image.open(image_file)
     elif image_file.startswith("data:"):
@@ -933,71 +937,6 @@ def monkey_patch_vllm_gguf_config():
     setattr(GGUFConfig, "get_quant_method", get_quant_method_with_embedding_replaced)
 
 
-def maybe_set_triton_cache_manager() -> None:
-    """Set environment variable to tell Triton to use a
-    custom cache manager"""
-    cache_manger = os.environ.get("TRITON_CACHE_MANAGER", None)
-    if cache_manger is None:
-        manager = "sglang.srt.utils:CustomCacheManager"
-        logger.debug("Setting Triton cache manager to: %s", manager)
-        os.environ["TRITON_CACHE_MANAGER"] = manager
-
-
-class CustomCacheManager(FileCacheManager):
-    # Adapted from: https://github.com/tdoublep/vllm/blob/3307522289fdfefe323b6c00d0db696651989a2f/vllm/triton_utils/custom_cache_manager.py
-    def __init__(self, key, override=False, dump=False):
-        from sglang.srt.distributed.parallel_state import get_tp_group
-
-        self.key = key
-        self.lock_path = None
-
-        try:
-            module_path = "triton.runtime.cache"
-            cache_module = importlib.import_module(module_path)
-
-            default_cache_dir = getattr(cache_module, "default_cache_dir", None)
-            default_dump_dir = getattr(cache_module, "default_dump_dir", None)
-            default_override_dir = getattr(cache_module, "default_override_dir", None)
-        except (ModuleNotFoundError, AttributeError) as e:
-            default_cache_dir = None
-            default_dump_dir = None
-            default_override_dir = None
-
-        if dump:
-            self.cache_dir = (
-                default_dump_dir()
-                if default_dump_dir is not None
-                else os.path.join(Path.home(), ".triton", "dump")
-            )
-            self.cache_dir = os.path.join(self.cache_dir, self.key)
-            self.lock_path = os.path.join(self.cache_dir, "lock")
-            os.makedirs(self.cache_dir, exist_ok=True)
-        elif override:
-            self.cache_dir = (
-                default_override_dir()
-                if default_override_dir is not None
-                else os.path.join(Path.home(), ".triton", "override")
-            )
-            self.cache_dir = os.path.join(self.cache_dir, self.key)
-        else:
-            # create cache directory if it doesn't exist
-            self.cache_dir = os.getenv("TRITON_CACHE_DIR", "").strip() or (
-                default_cache_dir()
-                if default_cache_dir is not None
-                else os.path.join(Path.home(), ".triton", "cache")
-            )
-            if self.cache_dir:
-                try:
-                    self.cache_dir = f"{self.cache_dir}_{get_tp_group().local_rank}"
-                except:
-                    self.cache_dir = f"{self.cache_dir}_{os.getpid()}"
-                self.cache_dir = os.path.join(self.cache_dir, self.key)
-                self.lock_path = os.path.join(self.cache_dir, "lock")
-                os.makedirs(self.cache_dir, exist_ok=True)
-            else:
-                raise RuntimeError("Could not create or locate cache dir")
-
-
 def set_ulimit(target_soft_limit=65535):
     # number of open files
     resource_type = resource.RLIMIT_NOFILE
@@ -2061,6 +2000,16 @@ def is_valid_ipv6_address(address: str) -> bool:
         return False
 
 
+def maybe_wrap_ipv6_address(address: str) -> str:
+    if is_valid_ipv6_address(address):
+        return f"[{address}]"
+    return address
+
+
+def format_tcp_address(ip: str, port: int) -> str:
+    return f"tcp://{maybe_wrap_ipv6_address(ip)}:{port}"
+
+
 def configure_ipv6(dist_init_addr):
     addr = dist_init_addr
     end = addr.find("]")

sglang/test/test_activation.py

@@ -3,9 +3,12 @@ import unittest
 
 import torch
 
-from sglang.srt.layers.activation import GeluAndMul
+from sglang.srt.layers.activation import GeluAndMul, QuickGELU
+from sglang.srt.utils import is_hip
 from sglang.test.test_utils import CustomTestCase
 
+_is_hip = is_hip()
+
 
 class TestGeluAndMul(CustomTestCase):
     DTYPES = [torch.half, torch.bfloat16]
@@ -52,5 +55,51 @@ class TestGeluAndMul(CustomTestCase):
                 self._run_gelu_and_mul_test(*params)
 
 
+class TestQuickGELU(CustomTestCase):
+    DTYPES = [torch.half, torch.bfloat16]
+    NUM_TOKENS = [7, 83, 2048]  # batch = sequence length
+    DIMS = [512, 4096, 5120, 13824]  # all multiples of 16 bytes
+    SEEDS = [0]
+
+    @classmethod
+    def setUpClass(cls):
+        if not torch.cuda.is_available():
+            raise unittest.SkipTest("CUDA is not available")
+        torch.set_default_device("cuda")
+
+    def _run_gelu_quick_test(self, n_tok: int, dim: int, dtype: torch.dtype, seed: int):
+        torch.manual_seed(seed)
+
+        layer = QuickGELU().to(dtype=dtype)
+
+        x = torch.randn(n_tok, dim, dtype=dtype, device="cuda")
+
+        with torch.inference_mode():
+            ref = layer.forward_native(x)  # x * sigmoid(1.702 * x), fp32 math
+            if _is_hip:
+                out = layer.forward_hip(x)  # 128-bit vectorised kernel from sgl-kernel
+            else:
+                out = layer.forward_cuda(x)
+
+        tol = 1e-2 if dtype is torch.bfloat16 else 1e-3
+        self.assertTrue(
+            torch.allclose(out, ref, atol=tol, rtol=tol),
+            msg=f"Mismatch @ B={n_tok}, D={dim}, dtype={dtype}",
+        )
+        print(f"Match @ B={n_tok}, D={dim}, dtype={dtype}")
+
+    def test_quick_gelu(self):
+        for params in itertools.product(
+            self.NUM_TOKENS, self.DIMS, self.DTYPES, self.SEEDS
+        ):
+            with self.subTest(
+                num_tokens=params[0],
+                dim=params[1],
+                dtype=params[2],
+                seed=params[3],
+            ):
+                self._run_gelu_quick_test(*params)
+
+
 if __name__ == "__main__":
     unittest.main(verbosity=2)

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.4.9.post3"
+__version__ = "0.4.9.post4"

{sglang-0.4.9.post3.dist-info → sglang-0.4.9.post4.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sglang
-Version: 0.4.9.post3
+Version: 0.4.9.post4
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -253,13 +253,13 @@ Requires-Dist: uvloop; extra == "runtime-common"
 Requires-Dist: xgrammar==0.1.21; extra == "runtime-common"
 Provides-Extra: srt
 Requires-Dist: sglang[runtime_common]; extra == "srt"
-Requires-Dist: sgl-kernel==0.2.6.post1; extra == "srt"
+Requires-Dist: sgl-kernel==0.2.7; extra == "srt"
 Requires-Dist: torch==2.7.1; extra == "srt"
 Requires-Dist: torchaudio==2.7.1; extra == "srt"
 Requires-Dist: torchvision==0.22.1; extra == "srt"
 Requires-Dist: cuda-python; extra == "srt"
 Requires-Dist: einops; extra == "srt"
-Requires-Dist: flashinfer_python==0.2.7.post1; extra == "srt"
+Requires-Dist: flashinfer_python==0.2.9rc1; extra == "srt"
 Provides-Extra: blackwell
 Requires-Dist: sglang[runtime_common]; extra == "blackwell"
 Requires-Dist: sgl-kernel; extra == "blackwell"
@@ -268,11 +268,11 @@ Requires-Dist: torchaudio==2.7.1; extra == "blackwell"
 Requires-Dist: torchvision==0.22.1; extra == "blackwell"
 Requires-Dist: cuda-python; extra == "blackwell"
 Requires-Dist: einops; extra == "blackwell"
-Requires-Dist: flashinfer_python==0.2.7.post1; extra == "blackwell"
+Requires-Dist: flashinfer_python==0.2.9rc1; extra == "blackwell"
 Provides-Extra: srt-hip
 Requires-Dist: sglang[runtime_common]; extra == "srt-hip"
 Requires-Dist: torch; extra == "srt-hip"
-Requires-Dist: petit_kernel; extra == "srt-hip"
+Requires-Dist: petit_kernel==0.0.2; extra == "srt-hip"
 Provides-Extra: srt-xpu
 Requires-Dist: sglang[runtime_common]; extra == "srt-xpu"
 Provides-Extra: srt-hpu

{sglang-0.4.9.post3.dist-info → sglang-0.4.9.post4.dist-info}/RECORD

@@ -10,7 +10,7 @@ sglang/global_config.py,sha256=xzLdk8W53fneFblNh8iIjGF9C3-7mnzR1-LleD9Btxg,1495
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
 sglang/profiler.py,sha256=tEHzHerXC-ymk4OrkoUcMbgcGHmb8VESthsNSP2Yx9w,4417
 sglang/utils.py,sha256=leKs-YyX1_Jk216yoKX-KRYx_EbLiuemsHbD31xSjMw,16445
-sglang/version.py,sha256=l7iWzcwTcItFAUU_b9YsWRVypWGj8cPQTjULXXXSKcI,28
+sglang/version.py,sha256=7hKPeJG6JhOmqx_uKkIvWkBNoTZDwS93b5EIfp7zQXQ,28
 sglang/eval/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
 sglang/eval/loogle_eval.py,sha256=-CC2s2kh5qUoDrHRkQVkC_jNvBgNojXbf456ny5s78s,4557
 sglang/lang/chat_template.py,sha256=HKlx7snSWFED8GKF5ex79sQrPWFw5TSXQM0_LsiD9Bc,20552
@@ -26,11 +26,11 @@ sglang/lang/backend/litellm.py,sha256=ugmL7sfUxkUHVbHtwNzHgdQAEd4UCjNQboFuE3KThc
 sglang/lang/backend/openai.py,sha256=YFs7pCDK6_DoYmjTa-vgDVdaeE4PYr7E-sIC1q7vAns,16422
 sglang/lang/backend/runtime_endpoint.py,sha256=oxLCLXCa26VfKyPV-wbRjLrQvqTG8NzoaVgwczzyIYc,17481
 sglang/lang/backend/vertexai.py,sha256=gz0uNYyBb88jbPYz6ZIJ774fefrcbuVdoK33bphUZpI,4827
-sglang/srt/_custom_ops.py,sha256=0lJRMTKTjoxJPh1qQnnMY02Z3SyBDi7LJI34IBLQsgQ,4462
+sglang/srt/_custom_ops.py,sha256=9GPNUyqnybgvUnzdQ8n5_NgluHmQNw0Gptos0iLPfrY,5342
 sglang/srt/aio_rwlock.py,sha256=6LYtOdeTUY3hkfa1dmYkgsaF2ttrwIF3hUWz2AZ2fqw,2970
 sglang/srt/code_completion_parser.py,sha256=KFa95OU0TeVjJkOqIgS4xV3kaJ5dFWOmAAgISyc1oEc,3803
 sglang/srt/constants.py,sha256=0i-tEwG2BSYNDy96MxnGHV5HnBELkYcnsVGsE-R18o0,93
-sglang/srt/conversation.py,sha256=_dDqfoULuZdDN41-3I-3oF5KVo35E9lv0KPjgAaG-mY,40399
+sglang/srt/conversation.py,sha256=8Ne6Wcjhdo9gSqIDNEcRCiPIIUawGjfDUimp5IHCeps,40396
 sglang/srt/custom_op.py,sha256=7OPQAZe3PqDXk-xM0Wxl7pm9cazilNJfkNiYIWztWgI,3198
 sglang/srt/debug_utils.py,sha256=slaFOY4BYDBFatkfu8FZlzai-u4LFS-5GUzdr-t50zE,2241
 sglang/srt/hf_transformers_utils.py,sha256=UiGjp70ZpTaZDJ99Roc-2DTkxV3gR-2oJjjaNY9Ekh4,12391
@@ -39,11 +39,11 @@ sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,
 sglang/srt/operations.py,sha256=ddQ8KO63L73OciaR8MZ9h2h83gKVY4-WuWgeEGowPJA,5346
 sglang/srt/operations_strategy.py,sha256=Pwd2sKeRtKh9WJXgzlNr2tU9y6YMcI3MDLbatHqlMws,7145
 sglang/srt/patch_torch.py,sha256=OUPCGQSQz3MVZB1zZ_Eq8lXiw0uIKJ_HWjqQolI8FsM,3088
-sglang/srt/reasoning_parser.py,sha256=DkZrFhkJ9qe7w46fTnM9sqhXwTDgZxD4qPdGka-Dly8,7579
-sglang/srt/server_args.py,sha256=2QFUu2xdPbIC_2Zf6BaBF_Ro9tSriUkvNp6P5VBZVOY,83290
+sglang/srt/reasoning_parser.py,sha256=8S7oLj5a1jPPwgs9Yl4sZ2A_wn_qobrmPjUJU0BswCQ,9224
+sglang/srt/server_args.py,sha256=Qdg567MVXoy2spYojSMY8MbcwFsALS7AnHCGXfqazAM,84056
 sglang/srt/torch_memory_saver_adapter.py,sha256=K_eTx0UU84MHSTXI3iqYLdHV4IWtJMJ2FKdGFJR8v1E,2417
-sglang/srt/two_batch_overlap.py,sha256=QxM1VzpAuiWniqsoHztDLTTfTsDVk4Lmh2pVq5flHxY,28610
-sglang/srt/utils.py,sha256=8QMSz7KDKbOuvtDLteVKLKOOyhmEhMYxvxBM670XPW4,92419
+sglang/srt/two_batch_overlap.py,sha256=Lx4bIXJb3hd9vzuJbmp3Crf4I8_NrT9fUbs-4LRzQXk,28711
+sglang/srt/utils.py,sha256=CSiP9vU3mxsAKQAK9DvgE4Ygd0LiHrl4zNJP5nuhqkU,89974
 sglang/srt/warmup.py,sha256=zldxhMlXpclRAJXmfBjJNUJd1eDizVdysibBvQyTVuA,1782
 sglang/srt/configs/__init__.py,sha256=8EcVRP95epZ49DxBa6LgKWt7eO3Qe7Hrr3V1c6HkMnY,553
 sglang/srt/configs/chatglm.py,sha256=j-b0YkdYUmQm2y1kNmMJtKeACxWKmBbvNNkDWbs6kbI,2907
@@ -56,7 +56,7 @@ sglang/srt/configs/janus_pro.py,sha256=yI2c_aWBIKiTcK6RZxK6hq8CIxyYRH_NN8uaSNKsr
 sglang/srt/configs/kimi_vl.py,sha256=4W7VQI3pr888ZsFA2SqCQo4mI0seXTOrGQ-x3oTvWew,1358
 sglang/srt/configs/kimi_vl_moonvit.py,sha256=hx2Rt4JSFbvy2HUTeLjBpge87m8M6ITAhqsgdNf_Jd4,1163
 sglang/srt/configs/load_config.py,sha256=qs-AxuplouBx2tsv9KGBOLZPbwzuVA4vbktbGP_cRp8,3309
-sglang/srt/configs/model_config.py,sha256=6X0aVKcWTJfu5Lw6gCtKodJVg7EUXcr2Js_NuGYeUYk,28401
+sglang/srt/configs/model_config.py,sha256=zLuzY5QlldWYuEb9zHlARTOuXYqZbXijhJHWMuy2lZA,28413
 sglang/srt/configs/update_config.py,sha256=D-myV9Ombwp6wTF0RX9qUO6m5rxFiUcyse32FQADG28,4787
 sglang/srt/configs/utils.py,sha256=3nHUfisMs_Ltuhv8OZTNCJp63YJKJVF43h1QZB1zqx8,670
 sglang/srt/connector/__init__.py,sha256=czLX5JOxuMhH-T9eSJzoc1qv1B4z9chyffDRL5I6wo4,1247
@@ -78,7 +78,7 @@ sglang/srt/disaggregation/decode.py,sha256=P-0OyFjjDfthEVhlJugdscWST_uoOVmHZlSlx
 sglang/srt/disaggregation/decode_schedule_batch_mixin.py,sha256=8UdNaj8KKMLV5Cydhw8gnHM-zRrnKM_AAd3Qc-SRfWg,5648
 sglang/srt/disaggregation/kv_events.py,sha256=yFbtwOeblcCmOXTjg00TopxnyjkmCBQIVz46KB8jetY,13555
 sglang/srt/disaggregation/launch_lb.py,sha256=mcbAztN4gnHevw_T5R2_nWsymsDEY9vHkm7OJ1vr6cc,4211
-sglang/srt/disaggregation/mini_lb.py,sha256=BBeIdeZZxi7Ra5_hukHHX3TNdyfZ4tP1GBUxC-qrv_g,14401
+sglang/srt/disaggregation/mini_lb.py,sha256=bpnz6d90nv7vAoiPxmxFKlM1Gjct0R4IPkF9SwwBk9I,14504
 sglang/srt/disaggregation/prefill.py,sha256=3u6WCrjogZSqd6yWdBZe1HO4lf6w2e9M7sQtGHyax-Y,23512
 sglang/srt/disaggregation/utils.py,sha256=ql5p9yjBCRcR0YMgcgRK6tH0kPGvLNTUgeXBr_BUcoI,12198
 sglang/srt/disaggregation/ascend/__init__.py,sha256=-lxnica6fZYmYNpKwmE8yESUpQ5Sxf2DiZoBHzboKc8,146
@@ -87,37 +87,38 @@ sglang/srt/disaggregation/ascend/transfer_engine.py,sha256=_4PszsQL6wICdg5cqpR2N
 sglang/srt/disaggregation/base/__init__.py,sha256=4VwUv0aWxwmVL1049XK82aLTNxmt0WY5RPy9li-wyVk,160
 sglang/srt/disaggregation/base/conn.py,sha256=CPDAoAkYaFtVPLa1QROfwipSVe7MH6omzIBHzo8TSYk,2811
 sglang/srt/disaggregation/common/__init__.py,sha256=7yl-EGLMVKRpBUaGF_7lwAsw2J_mqpRZV0238VGxD9o,126
-sglang/srt/disaggregation/common/conn.py,sha256=CZR1lMCE_mpSkvjc6BBmSr1SbHY1uZuxjpanazD-YXc,16071
+sglang/srt/disaggregation/common/conn.py,sha256=3YJVviObY3fkSfQbVxE5lgcqMmqM5nDjWfEjS5-SW6I,17054
 sglang/srt/disaggregation/common/utils.py,sha256=SxRhAWisNK8seGhb5BXBJ5u53DF7yeKVPMWPcB5ywbE,1194
 sglang/srt/disaggregation/fake/__init__.py,sha256=jJGWdXwaQiGIoR6atKqkQfkJmVyQ09l55VUN2WjwaeY,77
 sglang/srt/disaggregation/fake/conn.py,sha256=oD1DArn1yDFZCu-X6p93uSLlAXEkt9lYxERICMznxGw,2286
 sglang/srt/disaggregation/mooncake/__init__.py,sha256=0TgqkAdQI1YynbHY6c0QISvVoOSk-0SwCIq5rjPSmgE,156
-sglang/srt/disaggregation/mooncake/conn.py,sha256=ml3Ke6md9YYjG6yI4FAufF_642sbp2lOHVemF9Ww_Y0,60451
-sglang/srt/disaggregation/mooncake/transfer_engine.py,sha256=a3pki0d7QauTn6slt7gVxjicABP6Uo4Ztt0wEXR9dZc,4687
+sglang/srt/disaggregation/mooncake/conn.py,sha256=uHCmSIdo4-BfVez3uP4JmQ3vsRKXJhrcBAaRNi2dgBs,61314
+sglang/srt/disaggregation/mooncake/transfer_engine.py,sha256=-ulLhz6DBDsmEjwNjjH5Pj8ngZKISgwBc_zL3ViGDN0,4761
 sglang/srt/disaggregation/nixl/__init__.py,sha256=qODVPIGWUXKXq4zsRIcMYoAoAeg6nBIN9vdQOlVMANE,136
-sglang/srt/disaggregation/nixl/conn.py,sha256=cXqWU2Gbi6E2LS0MznEcgFCrwONhYxBtD-zE9DU32z0,20333
+sglang/srt/disaggregation/nixl/conn.py,sha256=eSof87fG21Dd4COszfnbeXIxne3TWvw0mSvCOkjsBZc,20323
 sglang/srt/distributed/__init__.py,sha256=jFOcyt-wFAPMBUAf9zkZalNQlt-4rqmT6pCKBz1E4qo,149
 sglang/srt/distributed/communication_op.py,sha256=IBnFUdMftK_VSTMMMitGveonorFUUVNL4guqO31cMSc,1130
-sglang/srt/distributed/parallel_state.py,sha256=nNLIRLFnLH4CtZqAt0CsuwyuTLgCrL5WjLDlS1-hj-4,56212
+sglang/srt/distributed/parallel_state.py,sha256=crlAUplgruIVrDk6jFZ0-6M3W6FXX67buZ3T-4XDgFA,58530
 sglang/srt/distributed/utils.py,sha256=aaCxATncLGnVgB0WlGpBdee0behKW8Dy_dakqcuKSaQ,8497
 sglang/srt/distributed/device_communicators/cuda_wrapper.py,sha256=3jvPG-Ow5UBLiXhfx8T8snR7crSZbPpARAggsDPWq7k,7038
-sglang/srt/distributed/device_communicators/custom_all_reduce.py,sha256=qje-PQ3v8yaV-oYVLPws1mgIlXVsGKFCOvXHmSe8ZXg,19624
-sglang/srt/distributed/device_communicators/custom_all_reduce_utils.py,sha256=fLoptT_U0lVAqkhEg-ge53CdFSIKQpDRiqHYKwJVEZg,10974
+sglang/srt/distributed/device_communicators/custom_all_reduce.py,sha256=p-GLYYnMSjcUvK30qcwYhXEJwGrOz6rR99p_23SWFj8,16570
+sglang/srt/distributed/device_communicators/custom_all_reduce_utils.py,sha256=Fx9D70NO2dplyuqOxTasMwkzONN3wfC4WOny3tWSbPA,14159
 sglang/srt/distributed/device_communicators/hpu_communicator.py,sha256=gPjEH1-izoby5uDrfUlzNf21luPT0Ow7pJjhCRKnHy8,1728
 sglang/srt/distributed/device_communicators/npu_communicator.py,sha256=bRXN1Md_4SHQGzQYZa2GrHv2zbIU5vSpkueHiAZL1xQ,1345
 sglang/srt/distributed/device_communicators/pymscclpp.py,sha256=8Pgehd02v-BpHixTTB4OB9ZlxA7fyXiPF4Xp9F_heyU,10890
 sglang/srt/distributed/device_communicators/pynccl.py,sha256=obXyCaZznZHSt486XCnEOBNG3Cen7ysuuMuGRlTTl-8,10095
 sglang/srt/distributed/device_communicators/pynccl_wrapper.py,sha256=LblisImY9d6EMz-oPS9J16WHo2Q_SRL1DtlJKK63Hfg,15349
-sglang/srt/distributed/device_communicators/shm_broadcast.py,sha256=d8mykYmXM1lfbPm8GNtqCF0Un_pdXYjbNmsgoVFyyow,20874
+sglang/srt/distributed/device_communicators/quick_all_reduce.py,sha256=4j1_E4azoxfd8wxtfFmt9rvbQncl8ny6wmTMl6gAkp0,9932
+sglang/srt/distributed/device_communicators/shm_broadcast.py,sha256=IrSrnpZnii0EJTS1CYRwEwE7gyHxJBVgI2QuJS3AKW0,20906
 sglang/srt/distributed/device_communicators/xpu_communicator.py,sha256=ajW6132BvA6jkeipEIgN27TFycI0U06Ih2Z8WNjlA4s,1593
 sglang/srt/entrypoints/EngineBase.py,sha256=yKN76witT2jz1zhmLHmPNLGMpK2UiOTaKQ2KPD8l99U,2594
-sglang/srt/entrypoints/engine.py,sha256=-J3rcOwCClya3h-jf6WjGDxLmJowJoZA4RVF9aRwwX0,31816
-sglang/srt/entrypoints/http_server.py,sha256=pUZgA7ILuRynIY3Cs2nbFoXSnZ7ohsMauF_RBtl-V_s,37538
+sglang/srt/entrypoints/engine.py,sha256=f8FI-BcjiIlcXdcy4NHmnM8SLNe8eYsNwyv4IoKYRZI,31545
+sglang/srt/entrypoints/http_server.py,sha256=-unKM0lCK7Scd3LLTxLl-3vhit6lMb-wWfjLywelRy0,37214
 sglang/srt/entrypoints/http_server_engine.py,sha256=_--j4U04OeJLlnnv1f0XmCd_Ry0z1FlhkrbePX8rYV0,4938
 sglang/srt/entrypoints/openai/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sglang/srt/entrypoints/openai/protocol.py,sha256=oAd5u_J6Wk9gzXVPS-Ps8VaxkhmMXyIuH1KZ73pEOvQ,19083
+sglang/srt/entrypoints/openai/protocol.py,sha256=UUc5oJ7E60fYoUqMb53H85i7HHhDfmYLIFZidhNrP0s,19147
 sglang/srt/entrypoints/openai/serving_base.py,sha256=5NJ2S_6B2NFSwn4nLp6eaeJ5iC3IcQzMEY9lW_gPcdA,5246
-sglang/srt/entrypoints/openai/serving_chat.py,sha256=o3-phSK6gHuG11eYpUbVJhz4Fq5yZeGRwQzIDhwdc10,37221
+sglang/srt/entrypoints/openai/serving_chat.py,sha256=WUstZZt80I2yrqUfBRx57-YMVMb45VqlyMnSmXuQFNw,38032
 sglang/srt/entrypoints/openai/serving_completions.py,sha256=emIutVmnJgOgGRuAqdB80-kVHUfQbE67n1VS-76o4QY,16690
 sglang/srt/entrypoints/openai/serving_embedding.py,sha256=hqPt2ELpT1yQ8sBXo801aLJ3sExoPXs_K5ZQtC2vpAs,6274
 sglang/srt/entrypoints/openai/serving_rerank.py,sha256=8n6y5kC6UhRpWrDtzH0eIp0dznW12FT60MLCR4bvD2U,3333
@@ -135,31 +136,31 @@ sglang/srt/eplb/eplb_algorithms/deepseek.py,sha256=mMZT7zAAArccdRS0xXxifvMb3qn9e
 sglang/srt/eplb/eplb_algorithms/deepseek_vec.py,sha256=Vzy5Iarua1VgVHzjBNZaVV_vt1LY1BLtJz7PmzB701k,10654
 sglang/srt/eplb/eplb_simulator/__init__.py,sha256=HIZaXqsvsOCMMJ81dSs1afSGZ5G6OiOZEtHpltyzzjY,21
 sglang/srt/eplb/eplb_simulator/reader.py,sha256=OlsWqsuXqVQY7PfvvrP-vsl2Ww2Kg4vD_K9BltM_QHo,1828
-sglang/srt/function_call/base_format_detector.py,sha256=eMSqbx99YPm2m_ChLafFRuLOzXBtA0oJn_Q4FtB3FnM,12950
+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/deepseekv3_detector.py,sha256=XudIiIt2Mejokg6qOkX7vYS-3XIFstOBHv-m-Q5fBLQ,8970
-sglang/srt/function_call/ebnf_composer.py,sha256=1r5UPEkUeMCug-P-c1mAZbJPgEXU8hR-fDLdYG5Q-wo,13585
-sglang/srt/function_call/function_call_parser.py,sha256=H0euMLfFJVPdKVjSJ0STsNdKEmdCfXExrug1LOXHxS4,7889
-sglang/srt/function_call/kimik2_detector.py,sha256=wb--wnAp-ZtWeIJZM1dS33GNu-K9lGxvrFbtp-UmGVk,8978
-sglang/srt/function_call/llama32_detector.py,sha256=EHYGAy1NCIxFWKdSQByPxyJgFaFRCtGdgj4ntEAgi00,3619
-sglang/srt/function_call/mistral_detector.py,sha256=xNuVl2vDXVYbXyiXLkJZ9VM6njcwlS9EqryR1eJbMr4,4462
-sglang/srt/function_call/pythonic_detector.py,sha256=rtXSflE4w993a4OqphWG-WQzvwRz0v3dfgHOnte1fpI,8731
-sglang/srt/function_call/qwen25_detector.py,sha256=9JfZem_5nw91Og2biwq8eIpUQjy_3kFz4TQI8Lc3Vow,4882
-sglang/srt/function_call/qwen3_detector.py,sha256=FWoXfa1dkvLT-zGmjxJUJm_BlCErZI23eWz3g7SwFP8,5361
+sglang/srt/function_call/deepseekv3_detector.py,sha256=WzJbwCe8DoOi8guX9pdVMd70CViVrwoaxi8PZwIC9pg,9712
+sglang/srt/function_call/ebnf_composer.py,sha256=cG0js5-bSl4FK66DyrTaREqLXXbtuLQxHBzxx6xaK3Q,14700
+sglang/srt/function_call/function_call_parser.py,sha256=G6b2VmE2mHarRuXlzAWEY6-yWoEbt830LmykE3Jnz9s,7904
+sglang/srt/function_call/kimik2_detector.py,sha256=7unW_GwH7I6jrh2BZbw22mWRZzQ0OMTYDz_xVX5abPU,10116
+sglang/srt/function_call/llama32_detector.py,sha256=trZja7IXc2IWUko5PRj7p4pRU6XCVjtxr-7qJNr0EWo,3647
+sglang/srt/function_call/mistral_detector.py,sha256=Ts2HrJdu5r2lCCMVKPZfnCds9pt1K5H1EaOQrF_HD60,4686
+sglang/srt/function_call/pythonic_detector.py,sha256=yAatItKtVDf9uhlB78wPWKP3fbefLJDNGGzFV5ey-YA,8756
+sglang/srt/function_call/qwen25_detector.py,sha256=nv83_sipZMU6Vie3J4HeNiZVCSRU5ei7601j50779HA,5181
+sglang/srt/function_call/qwen3_coder_detector.py,sha256=Zwpx4NF3dL7tOKnzczmvfmY3tZQ-vJyehbvtco-8tkM,5478
 sglang/srt/function_call/utils.py,sha256=__ImDF2kNyoLWsYO5RYoryvy1mmgEjnjXlCvLv-uLCM,1695
-sglang/srt/layers/activation.py,sha256=W1az1HEdUCnXblPORcMZHJwrxlYAb87v4JCIYF3xJak,7594
+sglang/srt/layers/activation.py,sha256=uhfhh10N5iLoLEBCvFPRjCfDxu5jXe1ehHJb3ISIjig,7840
 sglang/srt/layers/amx_utils.py,sha256=1mENgHK2B8mgaD1oMtgbZ15Jmy_Uu1QueBmo09Ff2iA,2865
-sglang/srt/layers/communicator.py,sha256=yckTsQ_3SeADES7maU2bgIpzwbWu3Q3Qpxu8T8UpCoQ,20045
-sglang/srt/layers/dp_attention.py,sha256=V7m_PfvHHtRMJMviIqmoyqf5VFFJLYYcynDuI5dXtAU,10275
+sglang/srt/layers/communicator.py,sha256=c5pJObNyP_7JsIWgLau_E-1ovtPCpAcge254fjo9Rqw,19988
+sglang/srt/layers/dp_attention.py,sha256=tOre7il5Cppu930-tzM6d4AgjHuIJMCMlg9hJkuBcjA,11382
 sglang/srt/layers/elementwise.py,sha256=MyQUflyKEfPZ-BggW1Kd4hB53RFD6FXGc2S5LXjx_do,16026
 sglang/srt/layers/flashinfer_comm_fusion.py,sha256=fkTcAB7qYwSWi95qI3Rqq0JUyDpJdcYkd4TYkWO01HI,5891
 sglang/srt/layers/layernorm.py,sha256=ooqA-t-vY5erbKBwqnOotfDsJRTygP5E10CfzEAVF6M,8657
 sglang/srt/layers/linear.py,sha256=NzjLqZrZNYcc3z2KtA9-n6pnSdBHHwU6NF_BbcAWp48,52471
-sglang/srt/layers/logits_processor.py,sha256=tOPMYopEQuoLBzwbjSmU9OWFYh2mmy3ViJSbA_wpFOU,25619
+sglang/srt/layers/logits_processor.py,sha256=9eZQE8pI-coXeK1XQFrBI0_aENZ2cGTtmD1FptWda5k,26270
 sglang/srt/layers/multimodal.py,sha256=YVR69WW-2aGDcZHT8IVJ6F_LRM7wraZr8VjrPDXqDmA,2104
 sglang/srt/layers/parameter.py,sha256=jCg0G-12GZqTa9gGOqEtVCsnnuBGnYg0VmF6pc6oCFg,17455
 sglang/srt/layers/pooler.py,sha256=uZ6WX1FLMEafZwusyZdm6KuVlIwSjbKrdwk2qzgqNGk,3812
-sglang/srt/layers/radix_attention.py,sha256=IlqRB4bk06FOH05_7zB8lik0xLpys7jFooLeCwdO0j8,3437
+sglang/srt/layers/radix_attention.py,sha256=vtzCPfEiWYunglcLfqO9dcATb-LXzU4sFBwIWsWHbiY,3513
 sglang/srt/layers/rotary_embedding.py,sha256=EhxI0E8jcTWZ2COpnku7crbW8Hew5fe_ujMndj7hKvE,52246
 sglang/srt/layers/sampler.py,sha256=xNds1migup2s6b9_pS6ljkJUkvNtv7nmTGeIdOzoQ6w,11182
 sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSahyMfA,4063
@@ -167,7 +168,7 @@ sglang/srt/layers/utils.py,sha256=IWGg1Hb7c33Z3LHRPVJyUAzp3BnSid23ZWXAmJ_Jvp8,12
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=FSAxHh1w9Qz4A2_8OQfIP5qsmq1s_g-Od-VLJ4vqQPw,22355
 sglang/srt/layers/attention/aiter_backend.py,sha256=7sEUgViw-xl3yok91yyOD9gTi8lQmME0g0ZiKVTCcyI,32851
 sglang/srt/layers/attention/ascend_backend.py,sha256=jPCsU9_gH1iZNoZHD9nCeDdVdXqBt31LI65N55BTJPg,8250
-sglang/srt/layers/attention/base_attn_backend.py,sha256=KXVcCguwXh-PSrY9Y2aUrlXXUhWdbVxqVEF2_xIMvm4,3466
+sglang/srt/layers/attention/base_attn_backend.py,sha256=_vM05ddWy6SaoqWYPR8w5lflSE4MhcVFR271Q9EKK8E,3600
 sglang/srt/layers/attention/cutlass_mla_backend.py,sha256=SIR7sKCCegwzahSz82I3gsDyN5TkKoa4yG4-pBQWBi4,9813
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=2ZRL_gYz14idoVqQzeQ6N77nXer0f_8_TUYw40XUUz0,9161
 sglang/srt/layers/attention/flashattention_backend.py,sha256=rYsBAsFS1qNNe-3CVIaXxWRqA3eul1jVQXvfZ_wlpwg,94076
@@ -192,13 +193,13 @@ sglang/srt/layers/moe/cutlass_moe_params.py,sha256=9NRCmgP_Ug3gGqCcpi-x-QRbLjCNp
 sglang/srt/layers/moe/cutlass_w4a8_moe.py,sha256=bc8s5Oc36pCF0VozxZuqFOOU2fov-4wGPlyduye3Imk,7296
 sglang/srt/layers/moe/fused_moe_native.py,sha256=SkOFgU23aZfHmLhnkVhdUWGx0-2wbo6uu9ohKBp2yOI,3486
 sglang/srt/layers/moe/router.py,sha256=UrPieRvemN7Ew48gtG7DA2xhNDBRSnZxzugTEBI-0_E,12006
-sglang/srt/layers/moe/topk.py,sha256=zjnI1s-NevasGpyuxurGVmCXMxGNfhp5F5onGePnf5A,25256
+sglang/srt/layers/moe/topk.py,sha256=bDHFcFQscnGDpCk7VejoChbpEVVz8Ph4FdzwcQRny0U,25047
 sglang/srt/layers/moe/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/srt/layers/moe/ep_moe/kernels.py,sha256=UFH-XacP3aBjAcn6_SeC641jGzjSrY0fz50IB9URix4,42706
 sglang/srt/layers/moe/ep_moe/layer.py,sha256=0vJW7RoEJ8Y3wD3q8pk289Gx5KNSVNJBbdpveliyXyM,46910
 sglang/srt/layers/moe/ep_moe/token_dispatcher.py,sha256=aiBE3mjvWV3eBrFGH9J44tuJncQwOjRS_XeyBNCEtqM,24379
 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=YytUs8ojQ83eyw9ObVZPxjdHXcuVbkEbpe1hLvI-nJQ,62950
+sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=6NcY-6UHQUSVP4qzc8jis4h-W9BEV4lnGmrJMx2S6eA,57164
 sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=BUbo3OFiSjqs34MK09McFYqSsKguDcG6KP1905WUMFA,23933
 sglang/srt/layers/moe/fused_moe_triton/triton_kernels_moe.py,sha256=hoWVdf8ry8IDGH2bfk2WW-y2S5h5haLTGanBSwkkeE0,5848
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_1_0/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=iNGsE2ZeVnQEnN4A8UJ9Jv0d3hbRF2MJ9oBgjup5Szk,2737
@@ -361,6 +362,7 @@ sglang/srt/layers/moe/fused_moe_triton/triton_kernels_moe.py,sha256=hoWVdf8ry8ID
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=288,N=64,device_name=NVIDIA_A800-SXM4-80GB.json",sha256=3T8_rF2PEojhgTMyQ8DscXgJCWWdWfDPj4M434zWcA4,3243
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_2_0/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3.json",sha256=tme0ydWzIxdABZLk4tU8G_X2dJUYGGZNkQzNGcmcvUc,3261
 "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",sha256=LNx1W_BsWZcpHomiScCRap46dV0-F7S_w3Htskoqlm8,3263
+"sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=160,N=320,device_name=NVIDIA_H20-3e.json",sha256=cWkEpNqnyn0QS8HcgWiwWI1xqh_U93_S1kyNeb69aOc,3238
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=257,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=fo1akCuGoqcGwQgfh56hApgg-wLXfo9kHHksE_6m1F4,3262
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ENDYnC1ljPNpDVHdmqR7UbgDWT1IP_55ZIQ_Oae34mI,3264
 "sglang/srt/layers/moe/fused_moe_triton/configs/triton_3_3_1/E=384,N=128,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=5UAs8Z1myw8F-5FY2HHWd5WMT2ii9X-aYIYkf3FsA8s,3269
@@ -381,20 +383,20 @@ sglang/srt/layers/quantization/int8_kernel.py,sha256=GfRC9FOn9exNvK4QHbUeBj3Hhv3
 sglang/srt/layers/quantization/int8_utils.py,sha256=YK9CS-lb_n91kNCTKK5o5apYF31V2giDg5G5VKrpcUA,2356
 sglang/srt/layers/quantization/kv_cache.py,sha256=_9pF5rwvB7ta6Gdc5YKVVGbNzYwqmhIx4TrX1-xnodQ,3261
 sglang/srt/layers/quantization/marlin_utils.py,sha256=ShsoxcVTYo1Es3XKp9mlQNw-TsRF_o4mHIO5MfAP5HA,25906
-sglang/srt/layers/quantization/modelopt_quant.py,sha256=V6FzVSNdqHr9nYe4667AeWdRTjY3n5Tp1d41W3eCYV8,37661
+sglang/srt/layers/quantization/modelopt_quant.py,sha256=ZxnHbYkpSt20xwV1zhZp3Xk93JzxBQOGHobZWIMluII,37872
 sglang/srt/layers/quantization/moe_wna16.py,sha256=NGcFjiMXwfOX10dkHAY3EgZHxox2k7I402Bvwj3jW1Q,19105
 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
 sglang/srt/layers/quantization/scalar_type.py,sha256=nBk4THjeRWMGW3hKWRdrjfQ0gwpuBxmYqapDBaRpnr4,12395
 sglang/srt/layers/quantization/unquant.py,sha256=hzyGGJoclb1oANy3hYxrEK76F7V7LNjU2rd3NWDpeGg,13997
-sglang/srt/layers/quantization/utils.py,sha256=mF5Hk42IHdGYi-l_1ZVHszXMY4WqgFEpOSDIlGZQ2sA,16469
+sglang/srt/layers/quantization/utils.py,sha256=9ZmfS0MMqueQ6zIxMIFoQy2OFBQ5vHjOHBgVvUn7ews,16213
 sglang/srt/layers/quantization/w4afp8.py,sha256=r36F5f5tAan8omW2UYZwK7xIwZUAWYHBbwJtmIIVzIY,9672
 sglang/srt/layers/quantization/w8a8_fp8.py,sha256=I_W9A4Pnluaf-SaT8zq-tAJTt1acpxfE_kG5mqdc5cE,10250
 sglang/srt/layers/quantization/w8a8_int8.py,sha256=VGrVELArNuyw4FEtUefaeK8ayuX7cpXj1OOqCmOCzm4,34902
 sglang/srt/layers/quantization/compressed_tensors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py,sha256=DsPCI82A4rqI6mjKo-WorIKUweppFV7-16Qku-zW9_g,25897
-sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py,sha256=Ss6bWLlmN9hNLbB9bdIaZFapiAJAzTaXhyf4pPwh1OY,24536
+sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py,sha256=vwcgwGuPhH2kHBfg7DnsndzffrbZx8xuFsl9Qeegd9s,24286
 sglang/srt/layers/quantization/compressed_tensors/utils.py,sha256=mnUmKWFQUnY8bVoFHUuNVwqsfS-cefeR-ofyaihCXcY,7621
 sglang/srt/layers/quantization/compressed_tensors/schemes/__init__.py,sha256=HWMTnmrj-mUCRXgcOwnnXLrvrAE-ONdPTSzSImjHCMA,347
 sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_scheme.py,sha256=tdKJC8c3SX8T3z8JL-1YCsg4ftcv55Wxt0vZrYftpX8,1635
@@ -559,8 +561,9 @@ sglang/srt/layers/quantization/deep_gemm_wrapper/entrypoint.py,sha256=On_uvP1Thx
 sglang/srt/lora/layers.py,sha256=FaTYf0w3zbKKKAT1nJdAzGRtxvySB6FpNKXvEi5RHH0,12958
 sglang/srt/lora/lora.py,sha256=sRoU7BdlTV4cnBv1-u3QjisVR9XQrrqAsmhxdNtd8dU,9433
 sglang/srt/lora/lora_config.py,sha256=qDgMTx_69jyJUl29O5FxLzYa0BMhqYVXWXfyyVOvGm0,1684
-sglang/srt/lora/lora_manager.py,sha256=biyPbUqnkez8OnSFXXjo_da8L0vH2ux954ypJTYfrIM,21667
-sglang/srt/lora/mem_pool.py,sha256=YsEJkddDhKDsCZ3f4hihhhZvyn0U1hWY07rx3zyrCPQ,12158
+sglang/srt/lora/lora_manager.py,sha256=ABuqg1IhY9LNXgJd_v3Yx_VHS52t4SapHmhLiH1thdc,20147
+sglang/srt/lora/lora_registry.py,sha256=q8HRG6YeWNKQFxbtaBlXSLGOgh0EtMAKPCKAsNEXdQQ,4747
+sglang/srt/lora/mem_pool.py,sha256=UYvTgEPvBoVsaX5rEnipebOAlo75pML6_NO9GfCFzl8,12148
 sglang/srt/lora/utils.py,sha256=5G0KiDEn-Zg4OgIOjWnU2ZdQGNstZswoAog9xU4GVDA,5660
 sglang/srt/lora/backend/base_backend.py,sha256=EIz8I-GIrdmK4fISw3ENhbJVVITaxKfyLxHXGPU4fPs,5044
 sglang/srt/lora/backend/flashinfer_backend.py,sha256=el6IAB4kTgDTbwCggmqFuukliyoapN5X6FLksG-4wJ8,4151
@@ -570,39 +573,39 @@ sglang/srt/lora/triton_ops/gate_up_lora_b.py,sha256=QURCYxHNR8Ls4SQtt3dvdgjvdDVh
 sglang/srt/lora/triton_ops/qkv_lora_b.py,sha256=iz5scFNo2YFTeiM8beWg9Z1oZI-6AM_T1wBMCQ6qp2Q,6485
 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=yJaVg78tM24CxVUk5-DDFXLiiLmqELyTpPUeW0CjlPo,22116
+sglang/srt/managers/cache_controller.py,sha256=-je55waMGlOXvEKICJ9CQCRngSwhdae2agFJCwO2E7o,24503
 sglang/srt/managers/configure_logging.py,sha256=8sNXZ2z9pBWOwn-X3wyz013Ob8Nbm1zDxRkxoZjH-l4,1633
 sglang/srt/managers/data_parallel_controller.py,sha256=PZ-wOcAFn3PQqMB6I2vjIsFmplf0nlNl5hqTDKTHTG8,12112
 sglang/srt/managers/detokenizer_manager.py,sha256=SpLxTsSPKBZfD-ZMhJ5zpPPGuUb8PmcYgFSL9CsurU4,10696
-sglang/srt/managers/io_struct.py,sha256=I8DQ348J2rz7LefwXINCdaq10vaBFrBi7kSDgpQRtHs,37205
+sglang/srt/managers/io_struct.py,sha256=LNLG7utMidF63eYWlWYxQ3G2GzOqaQzQc-9ld2yaycI,37859
 sglang/srt/managers/mm_utils.py,sha256=sHEOfM7T46btXHDArz5lS6pRL7gleAJpEz7l_z_MnWs,25476
 sglang/srt/managers/multimodal_processor.py,sha256=mzCrN-8H0bE0iMO8UzxmYmhE2M1qsbVJXGdhAYcjjYA,2016
-sglang/srt/managers/schedule_batch.py,sha256=y7BCydjHSn75zVuArzX6ecWM6HJD0rv70PXXsUTwfJ4,75603
+sglang/srt/managers/schedule_batch.py,sha256=GC0Pmz9eBI3da1zN0EkNvEQiOyB-qiouTcqITLx7peU,76068
 sglang/srt/managers/schedule_policy.py,sha256=BFYItMXtrsO5157tpFhYwLpZ8NSlzR4_gNZC1iE10D4,22400
-sglang/srt/managers/scheduler.py,sha256=6s0vJHsbKydAnH5E2e7TGr_sVFNTJH4GTbQRFZVLN0w,120941
+sglang/srt/managers/scheduler.py,sha256=OrWS4M1z68igMpFF-UROXXGec1PJOMBeUwqeOCLoBaI,120513
 sglang/srt/managers/scheduler_output_processor_mixin.py,sha256=Ei-9lyfrSQAlL9_ULuJLAhA7qpvUzjsBDzBBTDdnt4Q,31394
 sglang/srt/managers/session_controller.py,sha256=dzlMNZlo20FTSl64QqK7y7pElsdCy8ICOWWBPTBVwgs,6040
 sglang/srt/managers/template_manager.py,sha256=RrwRA2oqId_PMQ98qJQGwIxMroOxiorl2sGC9ARou_0,8543
-sglang/srt/managers/tokenizer_manager.py,sha256=XeP1jNaJtbLZ9tAMNEF5c9JUkUxuC6H9dwrrDT4gNt4,74743
-sglang/srt/managers/tp_worker.py,sha256=4W7rV7DsAViLygaWwtnQeR29ytfPkq1fnFpr3N1xPJk,11301
+sglang/srt/managers/tokenizer_manager.py,sha256=EIBoSgTGKuvYTy_hwTRvw69sbKex7jqPaFuE3P0lXqY,74848
+sglang/srt/managers/tp_worker.py,sha256=JEY1nJ6bx7IEYR-NlB2Ybuxw0dnD6N5G4XdFYOg4LQw,11257
 sglang/srt/managers/tp_worker_overlap_thread.py,sha256=kK0pn8mz-XgLGvVlJK6LvDVp4lnFk5vJH-Cs9NQ5Un4,10959
 sglang/srt/managers/utils.py,sha256=9_VGE48EK0PXVJ26aYvbRJ6n7gIZALvCcf6uZCccCgM,1369
-sglang/srt/mem_cache/allocator.py,sha256=drpOUu0aNCq-5R0rcQZCxyL8jwt2WFKRU255y2-gaZY,21081
+sglang/srt/mem_cache/allocator.py,sha256=id4riucsG9d-P1ikj3-DLJbzhM8-MuJfvDmB8H4g9-o,23364
 sglang/srt/mem_cache/base_prefix_cache.py,sha256=hLS2ncTMAz7Kpdk5pNwn5c6g8b61_K9OCBL4Vj1xsYc,2801
 sglang/srt/mem_cache/chunk_cache.py,sha256=FU_fq8e_dodgU8EkRqE4Jr0sKwCXQnKmapO88k6_JfU,3207
 sglang/srt/mem_cache/flush_cache.py,sha256=GYcxmNXh4hsMpFfNOuCTpKilW7guZwTtAg_usVeM3J0,979
-sglang/srt/mem_cache/hicache_storage.py,sha256=njuGCGuxO6ry-3Co2r41iRgKWM-J2nkUOD2xsuHwyPk,4968
-sglang/srt/mem_cache/hiradix_cache.py,sha256=76uxKFrgOO8yuSygqJhVyMMuPJk4qfpZ285bRoNfgTg,24224
+sglang/srt/mem_cache/hicache_storage.py,sha256=pvNeooVmCZBeZFgSHWNGPZzqtHI_qJ5BpTy_A4i6gg8,5539
+sglang/srt/mem_cache/hiradix_cache.py,sha256=1cx4yT4Or7HyvC1Qpi-thEfPoht9lLE9dcgWQf8SDWc,24758
 sglang/srt/mem_cache/memory_pool.py,sha256=jQdWgOr6hpMLbGaEJc4immBgZG_MZbx8LyOv029ASVw,40932
-sglang/srt/mem_cache/memory_pool_host.py,sha256=_FP-M8Cz2cRNQaaRRETM40PfPT1AmqT7DQHMU7cM7-Q,10255
+sglang/srt/mem_cache/memory_pool_host.py,sha256=xcjYyC94FOH_NguWbLUG4vF_CSBcxA_msuRsBasYCPo,10387
 sglang/srt/mem_cache/multimodal_cache.py,sha256=wZl2KeEl3xeoEsYdH33UoM-FO8kqfLo_XUgereJVvoM,1348
 sglang/srt/mem_cache/radix_cache.py,sha256=kp4JiVxh3eS4MsFxGKIYjYAb14B777qc1bHrABPqBMc,19101
 sglang/srt/mem_cache/swa_radix_cache.py,sha256=4qt_gS7GuHiekd6VvKkx54dB2GHJzNA_kj71Np6jTDA,40300
 sglang/srt/metrics/collector.py,sha256=FH6Wil2lMK7d7q74CfxfdULCaHKDXeAoo8UBZqqdepM,20289
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
-sglang/srt/model_executor/cuda_graph_runner.py,sha256=0f-50fkIBwNiy2gzf_c2ZvmGXtUgWA2GoYONx0MkXgY,30442
-sglang/srt/model_executor/forward_batch_info.py,sha256=jzhJFe7wTeovC2AgodaNUqH1dkY1ELJoLVcAyn-BjPQ,30290
-sglang/srt/model_executor/model_runner.py,sha256=oyqHt48lMMK9PBpT0inun1Sh3MOXIPr8ZD2iSwQb8O8,69663
+sglang/srt/model_executor/cuda_graph_runner.py,sha256=6njRGysrhBDi2bAt5DyjGnyrexYvuEVsAJwQBBW8Ut0,31886
+sglang/srt/model_executor/forward_batch_info.py,sha256=NvUhReevoy6UfO4991P_5MPQyfefC3BOoTwTqUwMbYk,38430
+sglang/srt/model_executor/model_runner.py,sha256=wVRFMda8oHYIeSfzINLlDMEQvff1uWMeJOwrS3U8_9Q,70162
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=wxSUYrqv23w9yHYy9Lm0ozWggs9pf6Co4yqe1UO8PZk,59799
 sglang/srt/model_loader/utils.py,sha256=svSrPQxViBz_-0_sByu4coCYrzJ51-4FCi_bFqa6rDo,4474
@@ -616,7 +619,7 @@ sglang/srt/models/dbrx.py,sha256=4pn_fdoATg01VEqNnIAxNEsKV5XU7gwHyd289eydq1s,155
 sglang/srt/models/deepseek.py,sha256=sqobTr9a9c5pNLNJZE-WrJIE_qBq0lC5gCGv9NpGU_Q,17364
 sglang/srt/models/deepseek_janus_pro.py,sha256=td8xGs6ARfJ8AQCYwUhMOZoWigrAs7m3trF5-kXCqik,70418
 sglang/srt/models/deepseek_nextn.py,sha256=47fehxRdiOizr0rdLg5f1fzQEx6gGAOcDcWKtblloyk,5928
-sglang/srt/models/deepseek_v2.py,sha256=g0SUVBZaw4dDK1waGN9fRoJNxnQ9Ughb25MJNQilVPo,104209
+sglang/srt/models/deepseek_v2.py,sha256=IG8hrn5j0o88mywclcyKe_djKLClC1Low6IDLNcQ23A,104131
 sglang/srt/models/deepseek_vl2.py,sha256=7X3gI4DE7guUtNJvMLf06jO8MUHKa2Aan8evZg5hsXc,13061
 sglang/srt/models/exaone.py,sha256=TpO-rtCpEZ8Ua7hGFnS8l2oAYhY0Pij50grc9WQ2mvc,13576
 sglang/srt/models/gemma.py,sha256=2atKwFN5wCsJCmBRmURX9vaCcFjxlFirL7xsnG5KPlI,14180
@@ -659,7 +662,7 @@ sglang/srt/models/mistral.py,sha256=ni7PppA_5X4ivTnIps3yoEaMEL-G6Gaf9b5hGC1vU5Q,
 sglang/srt/models/mixtral.py,sha256=j2whKt4GdQqtpKL5iN0Vb4TO4el1E2qEpUCI57E_47Q,17222
 sglang/srt/models/mixtral_quant.py,sha256=UUIq7lpUonD9IoGePyl59oJYEcVwT9wUXvtVp5tjQ9Y,15406
 sglang/srt/models/mllama.py,sha256=OdX7brhyfGiMFPulxF77QOUi-Kcscg3Qo4ipyxzs9yA,39670
-sglang/srt/models/mllama4.py,sha256=JsVrvDeWHwPemefvLMU8OWqm8uEQqE-n2CaOk8zoOQw,20237
+sglang/srt/models/mllama4.py,sha256=yt2YNE7S6buLAaqSfxD2LcJL1ZRzUzKWU21rCbzDzQo,20465
 sglang/srt/models/olmo.py,sha256=7-q_fA6XXdG7kPUjpUzYkzMUWJobuSjhqjYw9xSUs_c,12671
 sglang/srt/models/olmo2.py,sha256=azmljhJF4ivcQfUtfsAUxq3ducE4tRKTL6iwe0IKYMg,14327
 sglang/srt/models/olmoe.py,sha256=WdPtyKsd3u8OpVsXmDwWuVpt1gTF0679IRkc1pvfnMM,16040
@@ -677,11 +680,11 @@ sglang/srt/models/qwen2_5_vl.py,sha256=vOuIyK_NLuGgz9LlAdyM7cSQUU93VnBCYXUr_n4iO
 sglang/srt/models/qwen2_audio.py,sha256=lVeYmC29eVZIR9I6ZjFe4xvC0wSXR0NtGCli6fSXtms,7870
 sglang/srt/models/qwen2_classification.py,sha256=dGrMm4ebd30_lBhHOhaV57ig2iOTx3nqB4GEzsrRIM8,2747
 sglang/srt/models/qwen2_eagle.py,sha256=6spFg6JYQGeUXLv2Q5eMkVzZvnqGyvW6SdVBtFMxlTM,4806
-sglang/srt/models/qwen2_moe.py,sha256=FjR9y0w4kcDi8rZ3I8ai-2jf5osDUVNYYaSUKXKkjzs,25897
+sglang/srt/models/qwen2_moe.py,sha256=3OV_pR36OImR9U6H3hziSxDEIsPLKEAr_oO3hKzlNSc,25806
 sglang/srt/models/qwen2_rm.py,sha256=-mQXDEv11p-I1HXgYLTtY6ROem6UYorO958WsDrzsgs,2837
 sglang/srt/models/qwen2_vl.py,sha256=ZRAL32Ymt4ZCBqB0JorXAm4ZBrECZ9EbaXKaTSikkQM,21366
 sglang/srt/models/qwen3.py,sha256=COSC1TsCQNTq3E1QoVJYj1yTn-CFEVEX7keRwrrBXLs,20121
-sglang/srt/models/qwen3_moe.py,sha256=TbH1Y2-IcDhef5qjEogF1YXmTv2wsETQVuUytrO7eQo,34080
+sglang/srt/models/qwen3_moe.py,sha256=nSLBfzBzYQ6wpDmXPF2OV1v2kB1wTlnAV37LMj5guMQ,33911
 sglang/srt/models/registry.py,sha256=bH9H0OiNV0Cn7FRBM30Oh6dxICTs9mJscdLtHYPJvxc,3783
 sglang/srt/models/roberta.py,sha256=3k53V2Gbezk3jU-D03thXx8csGn5DxFK01ZQ9WFiWPI,9828
 sglang/srt/models/siglip.py,sha256=DIVJRwdtpLD2QT1kVPIHw7Bn9BE40xJbkHujDNzkjgA,9892
@@ -711,7 +714,7 @@ sglang/srt/multimodal/processors/qwen_audio.py,sha256=ILePOlBI5v8GI962q-pZ0dV-tk
 sglang/srt/multimodal/processors/qwen_vl.py,sha256=6maMXOwX3QICy3YNVSAF4VwxrZHCT4XH0MAAXIwIaeM,10464
 sglang/srt/multimodal/processors/vila.py,sha256=Dj66JpKbSdqpLNfk-NykYjhGAGuzBeMo7lFVCkeHkAM,2142
 sglang/srt/sampling/custom_logit_processor.py,sha256=AwODYVJdRkcQ8PGtJrhzKsqAgn8XZLQbAmR9fGiQzmc,1608
-sglang/srt/sampling/sampling_batch_info.py,sha256=m8m-mXu3N8hLYFhAkRopePDT5Ozkd4KAmWDgM_9_4iA,14536
+sglang/srt/sampling/sampling_batch_info.py,sha256=Duk4Fp99QZyOFld3i4AGwkW9FlULY86CNKv6v0_kA4w,14683
 sglang/srt/sampling/sampling_params.py,sha256=HbVcCeSXgPW19MRB-v0FTG1D-zHDAY6-toxyndSl-zI,6462
 sglang/srt/sampling/penaltylib/__init__.py,sha256=mtN8grFEcaBUhl4yBHmw8NNirt_i6uKO2cDNLHOpZQE,496
 sglang/srt/sampling/penaltylib/frequency_penalty.py,sha256=Loc3qjJTksNc5s-DV7QZHjgqoo5pxk7-nZzxwyhD2tQ,2144
@@ -719,10 +722,10 @@ sglang/srt/sampling/penaltylib/min_new_tokens.py,sha256=rdU_D7RoIcrQPhysNQEzmr4T
 sglang/srt/sampling/penaltylib/orchestrator.py,sha256=XM-Lm6u7gYPtMZrTIc0FR4QxNZxBH5s_Cj82umyCzYk,5721
 sglang/srt/sampling/penaltylib/presence_penalty.py,sha256=NRh10AJrrQlGJ6S-enGdRefrTrWpyqrSm-aNnyqQNQQ,2119
 sglang/srt/speculative/build_eagle_tree.py,sha256=O9LJNaBflgJdWT94D-rGH1gJFJ18nst2oOD8HnA2mZ4,12859
-sglang/srt/speculative/eagle_draft_cuda_graph_runner.py,sha256=fSNaHYJOQUgy2_AHRlZPYSQRIMgSiL_Pz5h53eothx4,14239
-sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py,sha256=LBAvhZ1ZUC-TSOQ45wG2GzBAc3kvSEdg5xKXS-xkBUc,15309
-sglang/srt/speculative/eagle_utils.py,sha256=cBQYqryCr4is5E_16xIcUue0Q2_Mn4meY8NCyuu4RAg,45558
-sglang/srt/speculative/eagle_worker.py,sha256=iOwloJqhYIrgBdnZQz9NV3zuiAe5dXJJa0QeSIg92x8,38180
+sglang/srt/speculative/eagle_draft_cuda_graph_runner.py,sha256=kVqMGj25jxtJlP-nTSlq_QJgoabFqKOU7-0WeUFhvw8,14509
+sglang/srt/speculative/eagle_draft_extend_cuda_graph_runner.py,sha256=hLjaD66K3ZE5SWsjvSiSdSexQ55F3EuSnfb4oe7rrSY,15261
+sglang/srt/speculative/eagle_utils.py,sha256=x5MKA2GkecNAXE58KKhPJRy5KxCLhRkD-C_kVSZIyzc,46433
+sglang/srt/speculative/eagle_worker.py,sha256=8yPQZxpBvcr--Ooo4nov3LqzkLRL-GOLJlwu91wMAmQ,38650
 sglang/srt/speculative/spec_info.py,sha256=rhaKG0TzyF9XZYHEWp1jccwTBohSNsUDvxHFtAoOl18,709
 sglang/test/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/test/few_shot_gsm8k.py,sha256=qnEejCyPHGEMBEiNc7T4-t9lvoo2DNN72RXw-ei_TnY,4300
@@ -736,7 +739,7 @@ sglang/test/simple_eval_humaneval.py,sha256=s8YuKIBkNUEOBP4XKsTMRdXGv5aRhl0oJyl-
 sglang/test/simple_eval_math.py,sha256=6kGKNwNbLN-Af3Wj8WTimWhH-Xp3enDmSvvSjsgWUpk,2550
 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=GeTIJHxlLQfW3kM-X1FGa8Sa3dSGKHEXl5wEy-hfGis,1489
+sglang/test/test_activation.py,sha256=bDBFl4Y30siXj7Ifl66fc2HytOrcZdHFiP2i9K8TIsY,3140
 sglang/test/test_block_fp8.py,sha256=hd3GXuC_wxn0HGAN3KCs6Nxf98AHO8XaRvGsvWS-CHc,21822
 sglang/test/test_block_fp8_deep_gemm_blackwell.py,sha256=Hnhq4kkyINHb4ONedkp5Kf7Xx0MjOoOBYL8J7ETvOVA,8180
 sglang/test/test_block_fp8_ep.py,sha256=XGZEs7cJiLjatRZcdcav4aFIQxnQB8Vq-b3HmAsw-Ww,10859
@@ -755,8 +758,8 @@ sglang/test/attention/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3h
 sglang/test/attention/test_flashattn_backend.py,sha256=_rTG849FwQdVTyGKkqhczaOqngBmRWXFmkl5NnuK1GM,13914
 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-0.4.9.post3.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.9.post3.dist-info/METADATA,sha256=QMSrhtk8BvwA8a_1DH3S77Yvhh6NCVv-KCef3au0MIE,27353
-sglang-0.4.9.post3.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-sglang-0.4.9.post3.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.9.post3.dist-info/RECORD,,
+sglang-0.4.9.post4.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.9.post4.dist-info/METADATA,sha256=moSWmWQPcs4bmIJ3FodeQnGZEf9NdVMbJLQlXXX4EAA,27348
+sglang-0.4.9.post4.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+sglang-0.4.9.post4.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.9.post4.dist-info/RECORD,,