sglang 0.4.5.post1__py3-none-any.whl → 0.4.5.post2__py3-none-any.whl

sglang/test/test_block_fp8.py

@@ -7,6 +7,7 @@ import torch
 from sglang.srt.layers.activation import SiluAndMul
 from sglang.srt.layers.moe.fused_moe_triton.fused_moe import fused_moe
 from sglang.srt.layers.quantization.fp8_kernel import (
+    per_tensor_quant_mla_deep_gemm_masked_fp8,
     per_tensor_quant_mla_fp8,
     per_token_group_quant_fp8,
     static_quant_fp8,
@@ -212,6 +213,62 @@ class TestPerTensorQuantMlaFP8(CustomTestCase):
                 self._per_tensor_quant_mla_fp8(*params)
 
 
+class TestPerTokenGroupQuantMlaDeepGemmMaskedFP8(CustomTestCase):
+    DTYPES = [torch.half, torch.bfloat16, torch.float32]
+    B = [128]
+    NUM_TOKENS = [7, 83, 2048, 1024 * 16]
+    D = [512, 128]
+    GROUP_SIZE = [128]
+    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 _per_token_group_quant_mla_deep_gemm_masked_fp8(
+        self, b, num_tokens, d, dtype, group_size, seed
+    ):
+        torch.manual_seed(seed)
+
+        x = torch.rand(b, num_tokens, d, dtype=dtype)
+
+        with torch.inference_mode():
+            ref_out, ref_scale = native_per_token_group_quant_fp8(x, group_size, 1e-12)
+            out, scale, _, _, _ = per_tensor_quant_mla_deep_gemm_masked_fp8(
+                x, group_size
+            )
+            out = out[:, :num_tokens, :]
+            scale = scale[:, :num_tokens, :]
+
+        self.assertTrue(
+            torch.allclose(
+                out.to(torch.float32), ref_out.to(torch.float32), rtol=0.20, atol=1e-2
+            )
+        )
+        self.assertTrue(torch.allclose(scale, ref_scale))
+
+    def test_per_token_group_quant_mla_deep_gemm_masked_fp8(self):
+        for params in itertools.product(
+            self.B,
+            self.NUM_TOKENS,
+            self.D,
+            self.DTYPES,
+            self.GROUP_SIZE,
+            self.SEEDS,
+        ):
+            with self.subTest(
+                b=params[0],
+                num_tokens=params[1],
+                d=params[2],
+                dtype=params[3],
+                group_size=params[4],
+                seed=params[5],
+            ):
+                self._per_token_group_quant_mla_deep_gemm_masked_fp8(*params)
+
+
 # For test
 def native_w8a8_block_fp8_matmul(A, B, As, Bs, block_size, output_dtype=torch.float16):
     """This function performs matrix multiplication with block-wise quantization using native torch.
@@ -485,5 +542,115 @@ class TestW8A8BlockFP8FusedMoE(CustomTestCase):
                 self._w8a8_block_fp8_fused_moe(*params)
 
 
+# For test
+def torch_w8a8_block_fp8_bmm(a, a_s, w, w_s, block_shape, out_dtype):
+    """This function performs bmm with block-wise quantization using native torch."""
+
+    B, N, _ = w.shape
+    _, M, _ = a.shape
+    out = torch.empty((B, M, N), dtype=out_dtype, device=a.device)
+
+    for i in range(B):
+        out[i] = native_w8a8_block_fp8_matmul(
+            a[i], w[i], a_s[i], w_s[i], block_shape, output_dtype=out_dtype
+        )
+
+    return out
+
+
+class TestW8A8BlockFP8BatchedDeepGemm(CustomTestCase):
+    DTYPES = [torch.bfloat16]
+    M = [1, 33, 64, 222, 8192]
+    N = [128, 512]
+    K = [128, 512]
+    BATCH = [128]
+    BLOCK_SIZE = [[128, 128]]
+    SEEDS = [0]
+
+    @classmethod
+    def setUpClass(cls):
+        if not torch.cuda.is_available():
+            raise unittest.SkipTest("CUDA is not available")
+        try:
+            import deep_gemm
+        except ImportError:
+            raise unittest.SkipTest("DeepGEMM is not available")
+        torch.set_default_device("cuda")
+
+    def _w8a8_block_fp8_batched_deep_gemm(self, M, N, K, B, block_size, dtype, seed):
+        torch.manual_seed(seed)
+        factor_for_scale = 1e-2
+        fp8_info = torch.finfo(torch.float8_e4m3fn)
+        fp8_max, fp8_min = fp8_info.max, fp8_info.min
+
+        a_fp32 = torch.randn((B, M, K), dtype=torch.float32) / 10
+        a = a_fp32.clamp(min=fp8_min, max=fp8_max).to(torch.float8_e4m3fn)
+
+        w_fp32 = (torch.rand((B, N, K), dtype=torch.float32) - 0.5) * 2 * fp8_max
+        w = w_fp32.clamp(min=fp8_min, max=fp8_max).to(torch.float8_e4m3fn)
+
+        block_n, block_k = block_size[0], block_size[1]
+        n_tiles_w = (N + block_n - 1) // block_n
+        k_tiles_w = (K + block_k - 1) // block_k
+
+        w_s = (
+            torch.rand((B, n_tiles_w, k_tiles_w), dtype=torch.float32)
+            * factor_for_scale
+        )
+        a_s = torch.rand((B, M, k_tiles_w), dtype=torch.float32) * factor_for_scale
+
+        ae = a.new_empty(B, (M + 255) // 256 * 256, K)
+        ae_s = a_s.new_empty(B, (M + 255) // 256 * 256, k_tiles_w)
+        oe = torch.empty((B, (M + 255) // 256 * 256, N), dtype=dtype)
+        ae[:, :M, :] = a
+        ae_s[:, :M, :] = a_s
+
+        masked_m = torch.full((B,), M, dtype=torch.int)
+        expected_m = M
+        lhs = (
+            ae,
+            ae_s,
+        )
+        rhs = (
+            w,
+            w_s,
+        )
+
+        from deep_gemm import m_grouped_gemm_fp8_fp8_bf16_nt_masked
+
+        with torch.inference_mode():
+            ref_out = torch_w8a8_block_fp8_bmm(a, a_s, w, w_s, block_size, dtype)
+            m_grouped_gemm_fp8_fp8_bf16_nt_masked(lhs, rhs, oe, masked_m, expected_m)
+            out = oe[:, :M, :]
+
+        self.assertTrue(
+            torch.mean(torch.abs(out.to(torch.float32) - ref_out.to(torch.float32)))
+            / torch.mean(torch.abs(ref_out.to(torch.float32)))
+            < 0.0001
+        )
+
+    def test_w8a8_block_fp8_batched_deep_gemm(self):
+
+        for params in itertools.product(
+            self.M,
+            self.N,
+            self.K,
+            self.BATCH,
+            self.BLOCK_SIZE,
+            self.DTYPES,
+            self.SEEDS,
+        ):
+            with self.subTest(
+                M=params[0],
+                N=params[1],
+                K=params[2],
+                B=params[3],
+                block_size=params[4],
+                dtype=params[5],
+                seed=params[6],
+            ):
+                self._w8a8_block_fp8_batched_deep_gemm(*params)
+
+
 if __name__ == "__main__":
     unittest.main(verbosity=2)

sglang/test/test_custom_ops.py

@@ -3,7 +3,7 @@
 import pytest
 import torch
 
-from sglang.srt.custom_op import scaled_fp8_quant
+from sglang.srt.layers.quantization.fp8_kernel import scaled_fp8_quant
 from sglang.srt.utils import is_cuda
 
 

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.4.5.post1"
+__version__ = "0.4.5.post2"

{sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: sglang
-Version: 0.4.5.post1
+Version: 0.4.5.post2
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -246,7 +246,7 @@ Requires-Dist: compressed-tensors; extra == "runtime-common"
 Requires-Dist: xgrammar==0.1.17; extra == "runtime-common"
 Provides-Extra: srt
 Requires-Dist: sglang[runtime_common]; extra == "srt"
-Requires-Dist: sgl-kernel==0.0.9.post1; extra == "srt"
+Requires-Dist: sgl-kernel==0.0.9.post2; extra == "srt"
 Requires-Dist: flashinfer_python==0.2.3; extra == "srt"
 Requires-Dist: torch==2.5.1; extra == "srt"
 Requires-Dist: torchvision==0.20.1; extra == "srt"
@@ -381,7 +381,7 @@ SGLang is a fast serving framework for large language models and vision language
 It makes your interaction with models faster and more controllable by co-designing the backend runtime and frontend language.
 The core features include:
 
-- **Fast Backend Runtime**: Provides efficient serving with RadixAttention for prefix caching, zero-overhead CPU scheduler, continuous batching, token attention (paged attention), speculative decoding, tensor parallelism, chunked prefill, structured outputs, and quantization (FP8/INT4/AWQ/GPTQ).
+- **Fast Backend Runtime**: Provides efficient serving with RadixAttention for prefix caching, zero-overhead CPU scheduler, continuous batching, token attention (paged attention), speculative decoding, tensor parallelism, chunked prefill, structured outputs, quantization (FP8/INT4/AWQ/GPTQ), and multi-lora batching.
 - **Flexible Frontend Language**: Offers an intuitive interface for programming LLM applications, including chained generation calls, advanced prompting, control flow, multi-modal inputs, parallelism, and external interactions.
 - **Extensive Model Support**: Supports a wide range of generative models (Llama, Gemma, Mistral, QWen, DeepSeek, LLaVA, etc.), embedding models (e5-mistral, gte, mcdse) and reward models (Skywork), with easy extensibility for integrating new models.
 - **Active Community**: SGLang is open-source and backed by an active community with industry adoption.

{sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/RECORD

@@ -1,44 +1,42 @@
-sglang/__init__.py,sha256=8J5PHcfRJul4R1NJnt0BtW05aVQNWqp8PyalLPOTZCA,1669
+sglang/__init__.py,sha256=T-fZEjKP66Q1q3PB56oREs5U3zf6bL0fNcdIbW8jMhE,1652
 sglang/api.py,sha256=vHiKBg8wwIdmrpnGclop5BzJ-1Q88emrlrfLwNCHg98,7010
 sglang/bench_offline_throughput.py,sha256=OQb-AjL4UNymmir02ht43uzgaNsnO_I11nXSowKMqBI,13841
-sglang/bench_one_batch.py,sha256=upXSMMlhO53J2cAP9lAb3KWt0nTEUdkNRWGk0lobS08,18887
+sglang/bench_one_batch.py,sha256=9-LFvhT0rjNa-Z5L0g5OpKfD6J4sQviRQbos42Fwkmc,18932
 sglang/bench_one_batch_server.py,sha256=8VYNhaQbWGP8TkNVuy_sPjD5FiuVZHamtGRWKwa-Z-Q,5962
-sglang/bench_serving.py,sha256=ek6D6uw0IlsMb0lhg57rBq7q4au7Os78GUMXopy0Wfk,57702
+sglang/bench_serving.py,sha256=6SnFF7rku8iDDW2qbIyqgBYsdZ3q3li5YH3zhhQH0E4,57540
 sglang/check_env.py,sha256=76itNLUw9KlqbiY1BI4u4YaMZaqyCNcrCLUIb6aHflM,8396
 sglang/global_config.py,sha256=xzLdk8W53fneFblNh8iIjGF9C3-7mnzR1-LleD9Btxg,1495
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
 sglang/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
 sglang/utils.py,sha256=GIcgiRHkZ-gyPxXOdn1qFF41jkg4-YdDxbPc4mzO-qk,16159
-sglang/version.py,sha256=mfIdQ0Yo6GF2VyWIDGyBUw_42D590eNsz05qnm3UXM4,28
-sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+sglang/version.py,sha256=0X83evIykSgmyAZc6vNVTXV2UQlVTx0omf8sSLruwLk,28
 sglang/lang/chat_template.py,sha256=MwNL5dNTe8g_l2ljZubnrazEgT2xEv-9O2D0Ezwxy4I,19658
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
-sglang/lang/compiler.py,sha256=o1C6G3TzhjSlsH-doTPy5oiVehr57dxNTa5oZw5TTAI,7639
+sglang/lang/compiler.py,sha256=MAuzoOOpb98njJ7Io2SDmFkhTroDYiq0te0ZpfHkMY4,7597
 sglang/lang/interpreter.py,sha256=OH1SFCm4rUCPO32MTo8j5V2Z13Jic7_r1GQOP1-aHaw,33234
 sglang/lang/ir.py,sha256=gssBGxqQEVJmjR_PqG2yah48AsInGaO3CmnQ2boJThc,18769
-sglang/lang/tracer.py,sha256=o-jLAPPSuy2vBfsGGrTAnbuWtORzQ50B4C_P5zvYkx8,8291
+sglang/lang/tracer.py,sha256=20B9w2W3d0D6xSUGZjKFszFuZAzsa3gvgdIgsXR8GtU,8176
 sglang/lang/backend/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sglang/lang/backend/anthropic.py,sha256=EXRX7xJgA5KZszX7toSLVnKzFQ5EO0Loj-YjHFtxSxg,2081
-sglang/lang/backend/base_backend.py,sha256=tdoh9YF3CyekY1BKiX9n7-aA4srDWIuA4RDJLM7q8qg,1985
+sglang/lang/backend/anthropic.py,sha256=qXWqqFlk0-nC62QCcYIRjLTc4AHrdTKpl7hCeDzROUs,2019
+sglang/lang/backend/base_backend.py,sha256=tEutGCU5tni1xGlrIjAVvA9aflJBt2duKrIDzqabJWc,1975
 sglang/lang/backend/litellm.py,sha256=ugmL7sfUxkUHVbHtwNzHgdQAEd4UCjNQboFuE3KThcY,2450
-sglang/lang/backend/openai.py,sha256=6I1udBC9obVpJG0GerbFVbWCbwLKkF2DoPdHBCnqPzs,16341
+sglang/lang/backend/openai.py,sha256=n47Zq1GTFc5x1DpN6CYL9cF6dfrsM7iJndnf5jRKU14,16331
 sglang/lang/backend/runtime_endpoint.py,sha256=CAVh3X9F80t_2tkJECF__7AdCQtqDg1AHDqIoKIPnvs,16755
-sglang/lang/backend/vertexai.py,sha256=O-iBLD-y3vq80UxnrAoJri7bxpgd-_eakZ88Cf8bEGA,4855
-sglang/srt/_custom_ops.py,sha256=lUBwC5R2UfjFMA1EtC5Kh2IngsqBJM9IuMW46kJWcjE,3647
+sglang/lang/backend/vertexai.py,sha256=gz0uNYyBb88jbPYz6ZIJ774fefrcbuVdoK33bphUZpI,4827
+sglang/srt/_custom_ops.py,sha256=L7NuEaRD_Q6Q54n0NZnLXgWZURbnn8Tkg4NQedE6zgA,3616
 sglang/srt/aio_rwlock.py,sha256=6LYtOdeTUY3hkfa1dmYkgsaF2ttrwIF3hUWz2AZ2fqw,2970
 sglang/srt/code_completion_parser.py,sha256=HhEUzdL-FVBsOot9tKDKA1l8Gdx8qsF1RRg-zHNpmLQ,5400
 sglang/srt/conversation.py,sha256=WP72AZrZpiqc5RowucT2tW3jVCb1pb4veW_kpwYS4yY,28785
-sglang/srt/custom_op.py,sha256=bIZ__3FiZvkbsN9O_jeLy_49X7ZbYbw0VxoL80uWwaI,3715
+sglang/srt/custom_op.py,sha256=J1PUcGaeJJjfAjp06BQsLpUkKyR1zsh9MvDiDlqqJsg,1129
 sglang/srt/function_call_parser.py,sha256=buYENeNEP5bhsvD424yGCa9wOqSfVOZSRn6zLiSJp5I,23733
 sglang/srt/hf_transformers_utils.py,sha256=N2f-gA8yUq-UP_TJT276gNbDNzmddWsmWnq3px6TIj8,9342
 sglang/srt/mm_utils.py,sha256=1ScBunw_x4W8ebM_AcJ62-1T2mfT8NlMJqdAhkF1lb0,12367
 sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,6078
 sglang/srt/patch_torch.py,sha256=OUPCGQSQz3MVZB1zZ_Eq8lXiw0uIKJ_HWjqQolI8FsM,3088
-sglang/srt/reasoning_parser.py,sha256=45xsU9RCPfyG4_Zx4y3-JPyNgAtrqwKI4j5R2NT4g1s,5594
-sglang/srt/server.py,sha256=PrQb9r6L9syWHKlggbbiQYsKtpwSmECqozRbf8qnoV8,874
-sglang/srt/server_args.py,sha256=k42YCDTbEEZZShmoaorQGNRwMxYACDSpvGW2toTb2DQ,53778
+sglang/srt/reasoning_parser.py,sha256=JnaEVW0KG1yJpn9uxmrjwErb9imzni05QDFjExryoqM,5584
+sglang/srt/server_args.py,sha256=2K5KQJgJ8T2q1XAnMIuIRTN5p5soFStsHPPb3n5yhJk,54193
 sglang/srt/torch_memory_saver_adapter.py,sha256=KG3wM9-xZsSdsmORofArnNR7hH55GEyFxaderCDcK9w,1853
-sglang/srt/utils.py,sha256=KAEwcTWJZInclD6tGvIEDcXpVW726l6n86FcbrHoCng,60992
+sglang/srt/utils.py,sha256=tbigwWXEsPOCD4rrOA4fzdZa4Qcf76GsagmgGW1DX4U,61565
 sglang/srt/warmup.py,sha256=FmJiYfjRr3X_eAe7ojQaPoN17LvHpjDmRWRnO-k86AQ,1469
 sglang/srt/configs/__init__.py,sha256=vulncVn70WqIT6s0HaB8p_Q6FjOiaLwNZWpoJS9FIuQ,399
 sglang/srt/configs/chatglm.py,sha256=j-b0YkdYUmQm2y1kNmMJtKeACxWKmBbvNNkDWbs6kbI,2907
@@ -61,18 +59,19 @@ sglang/srt/connector/serde/serde.py,sha256=n59I2MXLa7WCyN_8pEd8L-scJk7lMhmEX-GOU
 sglang/srt/constrained/base_grammar_backend.py,sha256=ljTVWpBo3bolce-E_-mtHIY2XWez4qcyDPeaIeZyIhM,7454
 sglang/srt/constrained/llguidance_backend.py,sha256=Kgd-PQVBQlKWsz506OpF_xSdNBhEbvFywzICTZg21iM,5729
 sglang/srt/constrained/outlines_backend.py,sha256=XbmkZSJzJnnY7k11uj8Et3StfuOiFwRs3ID4IRYAA4Q,6839
-sglang/srt/constrained/outlines_jump_forward.py,sha256=iZWXeR3gNYoMubLGyFmLPO4V2YsN5DiGjD71Xk9iFaE,6418
+sglang/srt/constrained/outlines_jump_forward.py,sha256=Gyubp-FVetxd6wP4FA_kD6cCXIRfr8k_ZDviJyte048,6824
 sglang/srt/constrained/reasoner_grammar_backend.py,sha256=XFxdZqvPofmtCeIMqR10NOyph06HwbdXfiVI8rIoV5s,3646
-sglang/srt/constrained/xgrammar_backend.py,sha256=Xf4CiU30XCa_RM4bgFkCw1yLeH4wijfewUHIYjnkv-k,6247
-sglang/srt/disaggregation/decode.py,sha256=qV0TU_nasIWTCXAngB7g3t5uJ1a8nmYW-KGXdIzaG0I,18624
-sglang/srt/disaggregation/mini_lb.py,sha256=ZU4M7ZtdKUDzpmNMT_NDFnTdbHzoGIxXSHW2PLdnoLs,7511
-sglang/srt/disaggregation/prefill.py,sha256=AeO4VcCKJ6X3c-GVY81G2aGA0bz9nNXiYNRXFOBWzWo,9954
-sglang/srt/disaggregation/utils.py,sha256=gbJIFpYM8XpW4aTThPGhny79jl9aBxOIiT2swJpS_Y8,2017
+sglang/srt/constrained/xgrammar_backend.py,sha256=pWyJL15D7kdiavcAG7_hMZEfSXbNbcxGE68qtrsLFsU,7288
+sglang/srt/constrained/triton_ops/bitmask_ops.py,sha256=WjTen9iuuFWLzkE1mAHQZB9_7aIy5QH8Wjf-lB-Fams,4614
+sglang/srt/disaggregation/decode.py,sha256=cB8Dc4eyH6BfRqwhgEyU3yvCw1XbFBaYlC9UaeMNERA,20948
+sglang/srt/disaggregation/mini_lb.py,sha256=7SJIcXPt-Fqpbs4dj93-4M30zFGY1AOEu6FCGlU1Uqg,7643
+sglang/srt/disaggregation/prefill.py,sha256=gm1VCWXv_s8iy2bR8RCYV1qbtdq8wW1dkoZmCURd3Fs,11453
+sglang/srt/disaggregation/utils.py,sha256=fXFdNQ9Mt-EnKqPfUpjG-f6bQhLKYxwKRppzWvWBBvY,2540
 sglang/srt/disaggregation/base/__init__.py,sha256=KR8xXoRCDAy2U623mfP6ujXu42m1_F9EiudjrKu2I_A,130
 sglang/srt/disaggregation/base/conn.py,sha256=gpf32bhYXWm_iaYB6WcrDaJ-UoL1ZzPI_xpi5pMhRQo,2443
 sglang/srt/disaggregation/mooncake/__init__.py,sha256=1vacEHmWjf7zgbMPzsXKB08FqNKNCquJdUiDlO41BOk,122
-sglang/srt/disaggregation/mooncake/conn.py,sha256=LkM9X7Rf3H4hjfSUQh4Yq7icvryTrPg_tszb3_hakcg,22210
-sglang/srt/disaggregation/mooncake/transfer_engine.py,sha256=qy-0HYLhaz90ompOtcOvB5jZhI97iAH1tl7dNvbleGc,3457
+sglang/srt/disaggregation/mooncake/conn.py,sha256=byeMgsi3Ape6sTNbF9KIvuq_R6FZMaCwEk5VoLqzeUg,21117
+sglang/srt/disaggregation/mooncake/transfer_engine.py,sha256=MxDAB9ZetRF1pFS2LP3FVHPtQ1HjIt_SK3UMaYHZ94o,2604
 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=hoTgLYfHIKMb_tSwBTauuusJZ8oY9BsiubTTOF8UfIw,50713
@@ -89,45 +88,45 @@ sglang/srt/entrypoints/EngineBase.py,sha256=xoyvp6XAeDLY2_Q2Ng33H-fRhrXHv2ldJJKd
 sglang/srt/entrypoints/engine.py,sha256=cO-Yq5i_hrn_yaAuhkHKkUUVXQmHXcTV4B-l76LjbwU,22137
 sglang/srt/entrypoints/http_server.py,sha256=wYjyyiajP6SWa3auZHZIUJv30zioB0IwdFKXHlyT5zo,28431
 sglang/srt/entrypoints/http_server_engine.py,sha256=ihA6y3GXRs28Y9U3SgdQcJQjnw_SVIby7QrVgiafX04,4846
-sglang/srt/entrypoints/verl_engine.py,sha256=sqQKt-HnEdCfR7CkiyskY7E_BeQMxASMDe_Hq91ni1I,6949
-sglang/srt/layers/activation.py,sha256=1ykXZO0BGz7DFVE-EK26b02I5AgH2IuU4PQB6oUcF4M,6003
+sglang/srt/entrypoints/verl_engine.py,sha256=XLYdwTwhH0jTjw8xczgZXWfBXMRb_ur2bg4TN0dTwfI,6975
+sglang/srt/layers/activation.py,sha256=pshilucJ66qSotOjvP7p3LmTBPVY6xBKkbwEpTLX_WY,5974
 sglang/srt/layers/dp_attention.py,sha256=Tfw2BydGPAeLcDlyl0jzBLOtkiygJhgAcf63RGZhrEE,7535
 sglang/srt/layers/elementwise.py,sha256=XCrR2i-9dP-H6jQo2zUuquwZrsl_wEQqj5Wxk6WUf7o,13987
-sglang/srt/layers/layernorm.py,sha256=189bORMggKhYcEYEjl6JRcuIoUPllHo3SheoH6YiORY,4546
-sglang/srt/layers/linear.py,sha256=etyzpgJbOUWCiu8abUovZQ8IFJ3ObeqOVHocBzFZOP0,51517
+sglang/srt/layers/layernorm.py,sha256=Z0N8S6vUX8F9SmPQi80pjaa6KGw5YJZYuVEqfgiZGV8,4546
+sglang/srt/layers/linear.py,sha256=nC9MxJrFap1BEyqgFlBySH4IeQruIbcBp32cOhUl5Fw,52149
 sglang/srt/layers/logits_processor.py,sha256=Vp8ibljVEezTr54xzeOcjiJR7JdYO8ItkO5nLIIMVu0,24206
-sglang/srt/layers/parameter.py,sha256=0OTMtmsNds42e3z3wHTRJiUfxCWFwSL6DHrqgeTgGt8,15151
+sglang/srt/layers/parameter.py,sha256=zqWyEzpWzP4NNTjq3G9khq6XofgpcmJqQLg6Vd4WyWE,15084
 sglang/srt/layers/pooler.py,sha256=rj2lygvleBnyLCBZ8I11HGMgpfIDsT0l3PIkshJwdu4,1606
-sglang/srt/layers/radix_attention.py,sha256=VeE8wX-8eNaSfw2JYzrek7EjgdVltEujnkMa_u92hCc,2865
-sglang/srt/layers/rotary_embedding.py,sha256=qvRNHWJxsyz897ntsEEr9iTj9ZNV_HgWUH7PvBE9ydw,45738
-sglang/srt/layers/sampler.py,sha256=yipSyN5UWGwGS-BC-WzWMmelys4CCDtK_8b1OpaK6sM,11622
+sglang/srt/layers/radix_attention.py,sha256=xcsEmKLZYu7K3RaYnMX67H86XLiKD8AGu0Ob-4oV8aU,3247
+sglang/srt/layers/rotary_embedding.py,sha256=-LXRtNCoawKiKwrtyhO91bMRBxKf8Vh06fF4P0zSJ0Y,45775
+sglang/srt/layers/sampler.py,sha256=pvtlV2L5RYiwsCr4A0tE1h_nou9grFbktFR8bcHMiLg,11404
 sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSahyMfA,4063
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=QUxd4sELx6p3dHvEKmccPZ-phdd_9EjNdwjH3SJ9zxI,22238
 sglang/srt/layers/attention/base_attn_backend.py,sha256=X_GIbQuU9njtUEGdUP7E_KRhmGxj3UyPHNESlL3QaQ8,3264
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=2ZRL_gYz14idoVqQzeQ6N77nXer0f_8_TUYw40XUUz0,9161
-sglang/srt/layers/attention/flashattention_backend.py,sha256=QKr_X_7fp96cbghJJj3qlzfN_ZeIZ-0a4wUMcSwwj-o,49327
-sglang/srt/layers/attention/flashinfer_backend.py,sha256=QqHbVoXp9LqoVvoGebXoo9GcrU7LaRRRh5sG93Daa0s,46277
+sglang/srt/layers/attention/flashattention_backend.py,sha256=9dv3cj7fRODFbc0v2pRLxmkFcSfLH5M-iVU_PpfaP3A,51364
+sglang/srt/layers/attention/flashinfer_backend.py,sha256=YtMTgMhxxNrAbSoWTPJczgY4SR3WjnAPXPoJ2d5PUZY,46394
 sglang/srt/layers/attention/flashinfer_mla_backend.py,sha256=pnVhvVEK87iFW8gUb1G7X7c1tqro8R2DSEOFCnlV8Bo,30301
 sglang/srt/layers/attention/flashmla_backend.py,sha256=1RPFNtQOBw6BWxIjrzfJgA9Nx92udLbR-S5KXmqjxS8,10536
-sglang/srt/layers/attention/torch_native_backend.py,sha256=KABmBrMqKa4x08kkQYdIcZUGydvmaVJIUfo3y8jhFHI,9270
-sglang/srt/layers/attention/triton_backend.py,sha256=cyxOaUU1CNhaEezJH9j0dd20cwxwIVGGN3jNXFTVkIY,26714
+sglang/srt/layers/attention/torch_native_backend.py,sha256=K5hUqBgakk2COSQqsaxWs0yEVOHS-7BlOygZTOeI8kE,9444
+sglang/srt/layers/attention/triton_backend.py,sha256=oEEiUwHbm4rNw5ExbQ2c3n0TwAgkk77yuLFenj9bHOo,26902
 sglang/srt/layers/attention/utils.py,sha256=J9mA-cbZT3uTlaKXo0HEAaeMei_TS2o4McTna9LVDCE,2750
 sglang/srt/layers/attention/vision.py,sha256=H7dQofAlZjC48Dr6bo4HBsZSLDhdrbHKChyGejeaEGo,11886
 sglang/srt/layers/attention/triton_ops/decode_attention.py,sha256=DPu_aCPgwPqKWZPEQmp_xA7MPbpV2ip-MEICCB470Ao,19120
 sglang/srt/layers/attention/triton_ops/double_sparsity_attention.py,sha256=BXUY8ARHBF2s9x9waiEwfZwcMgvuaJA0gxb4OeUZ_tY,31167
-sglang/srt/layers/attention/triton_ops/extend_attention.py,sha256=hbBvnhc2zqu-E3HNROVXyNOZbtDkVRuFus-yTjmE0Sg,13668
+sglang/srt/layers/attention/triton_ops/extend_attention.py,sha256=6MOrfQDVJQMTUXv0T3vUSF7JA67hVXxWqIbfDGBdKYM,13965
 sglang/srt/layers/attention/triton_ops/prefill_attention.py,sha256=Y66gZ37u0GKMPtI8n5MbO6uOxRuGEmKIG0IPbJTOqAM,6213
 sglang/srt/layers/attention/triton_ops/rocm_mla_decode_rope.py,sha256=664WnAJ91EiCUZOcnVDfbTQf4uGJ4ZDZB1CbxpEUFZc,13866
-sglang/srt/layers/moe/fused_moe_native.py,sha256=bf0po921lY9xnlZivdJly0bGIYFlLqp5v8Mz7tG5bdg,4451
+sglang/srt/layers/moe/fused_moe_native.py,sha256=IWoRLIEMhXH8KAB9jt5QSnhUtHn4SnbJ9dDO1LNTZ9o,4659
 sglang/srt/layers/moe/router.py,sha256=5Aeqoix_AS4uymb665OJE904wVSBkQeFdZP4e7KKPvg,10530
-sglang/srt/layers/moe/topk.py,sha256=xjkfC50rapfX27nI2078SoKGXAo82waQVArPVDH8Ehc,10843
+sglang/srt/layers/moe/topk.py,sha256=K-VU64nWBV07bu1Okn-uYbhz9gylq-KFNRYn2SFzu28,11129
 sglang/srt/layers/moe/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/srt/layers/moe/ep_moe/kernels.py,sha256=ijqRzS-tb0LGnDU5hW-g0JH104ppADrWaUIDGxb9Feo,22919
-sglang/srt/layers/moe/ep_moe/layer.py,sha256=1TmWnxv-bW1Qbgru-V-vGnt3ruuTIwHQy0Y5ZA_xzvE,36824
+sglang/srt/layers/moe/ep_moe/layer.py,sha256=Qotud38bCHg1tUtNKNjHPl4yHAD8ofELjtb8zcoePss,36350
 sglang/srt/layers/moe/ep_moe/token_dispatcher.py,sha256=zQV7Qr-Zrcr3D3efVvZepRQM02bj5djHPsijPssavk8,20430
 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=6WbcChGdZmSXl5_WlpC0w1cn_QH69mvICQt0pMBO_nk,55474
-sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=43-UL9KEMoaiC0cRSzWFbg2PADtcoxfZqjZ6TOvQ7Vk,24551
+sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=DRkGQ6vuk4d786_UWMBqYlw-9o75k-T_rdtLQLZldK4,55410
+sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=XofcceHnkmqhVczEIVqGSbsxps5LiYTGcaCk1Say9YM,24951
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=iNGsE2ZeVnQEnN4A8UJ9Jv0d3hbRF2MJ9oBgjup5Szk,2737
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=14336,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=JJN0hryyLr5Zv3dSS7C8cPFhAwTT6XxUVnBGMZvV6JA,2752
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=1,N=1792,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a16.json",sha256=ouRyZ5PEMPP2njPftCNhs-1g1y6wueWLmhI7G1SjV1k,4131
@@ -262,29 +261,29 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=43-UL9KEMoaiC0cRSzWFbg2PA
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json",sha256=-RzUWSIAAsg6iA-8SPMa68hPpBVoUyMJs3dLP7edRu0,4323
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=sY2nWMPh9lsIkhPCjkHO245wpnfFbrHmzdcZDVFPVww,3265
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=Uz5X80VcNBOaxshwVNUEittHk2zqB4HQCfTJ4TPG5aM,3274
-sglang/srt/layers/quantization/__init__.py,sha256=g5VWhdnQuyxFEcPyMhreR8QlEn3tXaCH34QPdAdufZQ,12369
+sglang/srt/layers/quantization/__init__.py,sha256=UOQcyCvKFkX0u_OPPex7X5X98iUR3lXgBnLbffu0n9g,12424
 sglang/srt/layers/quantization/awq.py,sha256=VImnVCU_QBLFba6S88T0dJ-vLy6SMm3OLIMEdllDfVI,6663
 sglang/srt/layers/quantization/base_config.py,sha256=jWk_egQrVNMYmQgbTI9vkcgzScLFjB5_sywFlAfE5J0,4776
-sglang/srt/layers/quantization/blockwise_int8.py,sha256=yE8ARplbha1sW1Szl-mgsRDzGTRpEZY_zAKkCJIu680,15010
-sglang/srt/layers/quantization/fp8.py,sha256=bYj6-xRO-bXsrDDaYzA2sKTmOvznLu7ZVoAtGR7cKjM,41834
-sglang/srt/layers/quantization/fp8_kernel.py,sha256=F49gP48suKwzO1QejmGSV6XrBSOgwW-bsaM-rvUB_mE,27460
-sglang/srt/layers/quantization/fp8_utils.py,sha256=ymdhxw-vMfJIzxW0uFg_iDTzlurN_R9dFXyEmYXsluI,22516
+sglang/srt/layers/quantization/blockwise_int8.py,sha256=cu9-JiCZDfMfvB97Kv_-eEG87VX5bRFIllFkzpO_xIg,15122
+sglang/srt/layers/quantization/fp8.py,sha256=Niu89OfZd4pIpkwZ1zd2Hrlffx0c5L5zkix6Lzi0Vys,40970
+sglang/srt/layers/quantization/fp8_kernel.py,sha256=oYq-j5moiilrTndKTqF4mRxCvhB1nSYhPMC9VeqcR4w,32964
+sglang/srt/layers/quantization/fp8_utils.py,sha256=LwP0Z8jPlaWpGBKVd27xqSk0ATzuB9CeLNEXiPm8wXs,18583
 sglang/srt/layers/quantization/gptq.py,sha256=e4rMz374-yQQqeAI77WPxfcAaRk38GeN2akEpvnC_Do,15141
 sglang/srt/layers/quantization/int8_kernel.py,sha256=GfRn_imIw8kNgqdtb2lr7BettjgDgimbl1Rubnamjh8,11352
 sglang/srt/layers/quantization/int8_utils.py,sha256=YK9CS-lb_n91kNCTKK5o5apYF31V2giDg5G5VKrpcUA,2356
 sglang/srt/layers/quantization/kv_cache.py,sha256=-yaFTdB75T0BbvQeuIpH6rZoL3R8t6OIJVGB-xdtpCw,3492
 sglang/srt/layers/quantization/modelopt_quant.py,sha256=Ff7qMv7CCWj0QY5gkDnwlQYLH1mbMtopbw8jXMLndXg,16616
-sglang/srt/layers/quantization/moe_wna16.py,sha256=3Z8Eq4_ehTN5EEotlYC09FpUNmF8VO8uv7QzUqJa0QI,19371
-sglang/srt/layers/quantization/utils.py,sha256=QqGFwRnFenOm5HfyLoS4D06_LyvNWgOggAiFtZXTpQ4,5637
-sglang/srt/layers/quantization/w8a8_fp8.py,sha256=wY9Ztw9RM5Vd3MHMLauy0KD8xcQ8JZUB_M4LeyE8-UU,11654
-sglang/srt/layers/quantization/w8a8_int8.py,sha256=fIoSGeaL5kZNrExKAKWBgJc9hNXns_w9zP7vw2dVPHA,8892
+sglang/srt/layers/quantization/moe_wna16.py,sha256=KtFr4lIslMA12yx4JjXXPOsa5OHjxXWA6scYCRQnFMQ,19483
+sglang/srt/layers/quantization/utils.py,sha256=3fP11UCSWkFWW7oTfQ6_3I1ZXfHvRL4WIlTAXnT1Ues,5442
+sglang/srt/layers/quantization/w8a8_fp8.py,sha256=VhM36MKz02W3uPCi-9Ap0XpQPXBdL88ny3l_aEtUq2M,11766
+sglang/srt/layers/quantization/w8a8_int8.py,sha256=_oOJMkMrRMU2WvzAsNsQubQaVYcJJx9kr7Pps5MW2dw,9002
 sglang/srt/layers/quantization/compressed_tensors/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py,sha256=Z2h9WQ74Umun_0wj8I5fr-ScRfuSrpSU9otI29jLAM8,25606
-sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py,sha256=RYX3vJ0Dh984KktTkNr1Zt578W9gp-YFY5nDiOpd7IQ,27073
+sglang/srt/layers/quantization/compressed_tensors/compressed_tensors.py,sha256=ssrSfCJ9ORpxsXNfCiKioxmrwY0alPTPd52YLEqiqlk,25634
+sglang/srt/layers/quantization/compressed_tensors/compressed_tensors_moe.py,sha256=no7gs-M8eEYvNd0XPoVudfb1mBweoSFfcHYoWytJeAY,26199
 sglang/srt/layers/quantization/compressed_tensors/utils.py,sha256=mnUmKWFQUnY8bVoFHUuNVwqsfS-cefeR-ofyaihCXcY,7621
 sglang/srt/layers/quantization/compressed_tensors/schemes/__init__.py,sha256=qcdRgoUNJWXqSimns-D987TW0OTk3uFuWNBX8Z6H8Fk,246
 sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_scheme.py,sha256=tdKJC8c3SX8T3z8JL-1YCsg4ftcv55Wxt0vZrYftpX8,1635
-sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py,sha256=BSgSFFeuBHSnZxvVpNDojeGCx0ClDDH04aTjdBUyE6U,5940
+sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a8_fp8.py,sha256=NZurhURFpZKqfMfgyd7oHLTLThm_8AO7xBCY8F6i3Gk,5881
 "sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_A100-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json",sha256=RdHQxWXwXqvio31192vsLaKjEr4f_DjpMPKlarY1IAk,3251
 "sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_A800-SXM4-80GB,dtype=int8_w8a8,block_shape=[128, 128].json",sha256=0vLaJgo5B9ti-XMFKJuvSoMGjsZQ-RhHSx4cC8Xji-U,3254
 "sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=tkLjwLC_aVXhzuvo-2QHkojXZauPJsf3jNHFn1S7uRA,3244
@@ -437,16 +436,15 @@ sglang/srt/layers/quantization/compressed_tensors/schemes/compressed_tensors_w8a
 "sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=PD4AJYCkHfy2ivv9baMouFXzBTy0eKMumbAfxfm91HI,3256
 "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/lora/layers.py,sha256=eqU5PxLx9jsmp0fxQ-e9mlSD0Zz4Y9Uan_x9Z5-y1EQ,11835
-sglang/srt/lora/lora.py,sha256=9BpasJObx-XL_qSoQhheGaYBJ3PlKqJAARvruIOJE0w,7514
+sglang/srt/lora/layers.py,sha256=cu1kqDCuH05ck8HVtwmVuMVBzcPJZeDY3mk486teB4E,11848
+sglang/srt/lora/lora.py,sha256=uNvbjZ_Wr1SLI9-ElRJA_JKwkibSGroP5Bfpsr9MI-Y,7527
 sglang/srt/lora/lora_config.py,sha256=qDgMTx_69jyJUl29O5FxLzYa0BMhqYVXWXfyyVOvGm0,1684
-sglang/srt/lora/lora_manager.py,sha256=Gpkq4N_cJGMIDtxUCScwP4LGcHyUJZ457EI_ti30_A8,9187
+sglang/srt/lora/lora_manager.py,sha256=nyqkm7RLoQE6myfqcH9r0zwME4aEZ3pFkVjY36QTlvA,9200
 sglang/srt/lora/mem_pool.py,sha256=xUFoHUDJgX9lt2YugD9HUY5tIMnJiazYMZ6LYqSGv-E,9633
 sglang/srt/lora/utils.py,sha256=GjEBgsGhDhX4NqVqeaciznQ8RotKZmb2c-nw4YMLHxA,5251
-sglang/srt/lora/backend/__init__.py,sha256=FziFT8HguMFj-h0tUCc4_UEbtOWMlYi4gNlYJcArWh4,671
-sglang/srt/lora/backend/base_backend.py,sha256=tGpABAn3DVC8GONf8USkaxkzkpVsDYfgKrnLCsXpivo,4558
-sglang/srt/lora/backend/flashinfer_backend.py,sha256=VmDSY2YqTLK2EBeqdMiNoirPxDifCMmfiCB3HNwpgvE,4138
-sglang/srt/lora/backend/triton_backend.py,sha256=_QbqggFPHMPWgx3PI15yyyfxBCTxSpGA209x_frI12E,2517
+sglang/srt/lora/backend/base_backend.py,sha256=EIz8I-GIrdmK4fISw3ENhbJVVITaxKfyLxHXGPU4fPs,5044
+sglang/srt/lora/backend/flashinfer_backend.py,sha256=el6IAB4kTgDTbwCggmqFuukliyoapN5X6FLksG-4wJ8,4151
+sglang/srt/lora/backend/triton_backend.py,sha256=uqwBGlguXX8EkCKjSPqac6SPTgcKA31u3u6HkKPQcos,2530
 sglang/srt/lora/triton_ops/__init__.py,sha256=JGOYPIn1XbGcyJTbt8A0qoc02PYONSGNNjGkC8yJpAM,283
 sglang/srt/lora/triton_ops/gate_up_lora_b.py,sha256=CDGt7lpu9GjykgMtmwbZ3PEqjTlRYyh28AUlj1cRcmw,5279
 sglang/srt/lora/triton_ops/qkv_lora_b.py,sha256=HTfU3HxxxVyaG_aJrrVjPJTnqf62yvepcKJKYkG0XJQ,5944
@@ -455,21 +453,21 @@ sglang/srt/lora/triton_ops/sgemm_lora_b.py,sha256=Q58UzWUb3QFqY_ZxWA3poN373N0Hwk
 sglang/srt/managers/cache_controller.py,sha256=d4RGqbut1FlzJnpqr7WY_TYmRjYPS07OoOVbztjs5xI,18959
 sglang/srt/managers/configure_logging.py,sha256=fOJaXAQ1n9m-8KPJndpsKvS885i69SMafoEADLIVfIM,1633
 sglang/srt/managers/data_parallel_controller.py,sha256=Xkj2n9uDyq7a-AVDZlfzeuNkC4ibsSftb1_bed9hgQ4,10318
-sglang/srt/managers/detokenizer_manager.py,sha256=HTfpJWMF1EImhKOnLJ96xPmYXm71xzaisLMfxg3zpgs,10111
+sglang/srt/managers/detokenizer_manager.py,sha256=3S3aRvKSi75RQSxEEQkeyxKDNNunWiw9wlwsbT1VXSo,10099
 sglang/srt/managers/expert_distribution.py,sha256=r3o5RGI0gnV7xb60AApqKYa0oiSB37oB7hQBX7P3xZM,3225
-sglang/srt/managers/io_struct.py,sha256=88tlo4xLTqpb_qv2mJpJPZQxpcWCK3xW34GUij05McQ,30860
-sglang/srt/managers/mm_utils.py,sha256=BdeiJG1vR89v8j9NwdHdanZlX5iiyxYWwHL5T0CjwUg,15537
-sglang/srt/managers/multimodal_processor.py,sha256=37SSZIdhdmcGaZSH2A2GLdntcbIxDUiomX6WR_BpmtQ,2132
-sglang/srt/managers/schedule_batch.py,sha256=4Xi7SdCiTKWraU8xQJQT1XYGMNt8TUW0dsZaNMDLutM,62509
+sglang/srt/managers/io_struct.py,sha256=nNRAJXJpJvZFHFkOPHVa9TkzawlhZlfHpu4qlZsYPWw,30897
+sglang/srt/managers/mm_utils.py,sha256=P-UbCdzasK0yR-xn6Lrk3ILct3uJrIBuXSDrtRZlB_A,15542
+sglang/srt/managers/multimodal_processor.py,sha256=XlRYvNhF6XOssreRX9DZPhLSpps_VE62gSKw3EGdNPo,2088
+sglang/srt/managers/schedule_batch.py,sha256=pPKw8z7B_6tA9OZoMU2r_KPJzAx3aNjylZx2Hm1bEcw,62407
 sglang/srt/managers/schedule_policy.py,sha256=E1qVq2G3jptKdX9nlqfayeRBUll9xB6bK8nBf3EW32E,19469
-sglang/srt/managers/scheduler.py,sha256=89db4YxUDZyrEZAKdbrxNIWk3EPob2aqvtZK4G8E2D0,82318
+sglang/srt/managers/scheduler.py,sha256=SujQHQOzw-O1NDxgMF8JWI0nF-EdRDC2ynKOXp69cNo,79876
 sglang/srt/managers/scheduler_output_processor_mixin.py,sha256=u2sj6MViFTov0lVZSysZ-wph2pEqRCtCjwA1UdttZ7I,26338
 sglang/srt/managers/session_controller.py,sha256=o-ifit0n4_xHLNmyD0Ams8FxGRgxFybX-Vz1hwgr3UQ,5755
-sglang/srt/managers/tokenizer_manager.py,sha256=ncPzDWeTgqi--V2LgRrzVIGp2aPl9Dcsv3qWCLDTBE4,46665
+sglang/srt/managers/tokenizer_manager.py,sha256=ukKGBjsVdI-FS3NF3sywxAHpWvK_eEOCDsfI0mPPPqw,46706
 sglang/srt/managers/tp_worker.py,sha256=khF-hXOrtF_IesOyUSjEBjb7fAh3CakdiKR7Ebj2wp8,8894
 sglang/srt/managers/tp_worker_overlap_thread.py,sha256=3_ZJ8Rq7v2ZDaRNTRu5Dy8AbqiAlJQp3IAKnn_WAwd8,9127
 sglang/srt/managers/utils.py,sha256=5i75uLlQOF_5CaT02CrWtwozMTtwTg2_nLP8Dtr-JZQ,1536
-sglang/srt/managers/multimodal_processors/base_processor.py,sha256=5pkKBqajyBRv7uM183NnrhTMYybC0HeUzHID_OkEnfA,9859
+sglang/srt/managers/multimodal_processors/base_processor.py,sha256=M36x_Emm-IdpFiQpii3qyQZ79O7NZsyJgDT5gk08Plo,9923
 sglang/srt/managers/multimodal_processors/clip.py,sha256=lRc2mcuDbAhZVf-0EfkO81pqDiol9zLvTpDqtPIBQ2k,1525
 sglang/srt/managers/multimodal_processors/deepseek_vl_v2.py,sha256=j7j1D38azudJjYthVpdz7jxQ9Z7SjwQfskpOIshAdiY,3147
 sglang/srt/managers/multimodal_processors/gemma3.py,sha256=UlkyIoc8XOw69iFBYiBYLx--pdfnM4JfCFtwRrd3w-o,2267
@@ -482,28 +480,29 @@ sglang/srt/managers/multimodal_processors/qwen_vl.py,sha256=67EmFiAkvZncU-eqiiS0
 sglang/srt/mem_cache/base_prefix_cache.py,sha256=NY62Zo0A0tLJ7ObRLOQqQcXCxoJUDZsK8f5U4dNQjKc,973
 sglang/srt/mem_cache/chunk_cache.py,sha256=it5SfL1FwMbrdeOH-I-Eu_i-I9hFB1xL-z_brIUoCkk,1835
 sglang/srt/mem_cache/flush_cache.py,sha256=GYcxmNXh4hsMpFfNOuCTpKilW7guZwTtAg_usVeM3J0,979
-sglang/srt/mem_cache/hiradix_cache.py,sha256=t3zxOCg8A4uMdjrtKbSdDJBwFubqnhfGOEdSs_22Zb4,16161
-sglang/srt/mem_cache/memory_pool.py,sha256=MyFVt81pCiiHoO3zRbIP_Z-KSAhHeSQGFQuL_kFL_L4,31638
+sglang/srt/mem_cache/hiradix_cache.py,sha256=hCexDvC_BC5juQHAuzZRE4e9Stirq1ysfsB87uz5BcQ,16340
+sglang/srt/mem_cache/memory_pool.py,sha256=bMzNK-8avd-_rwKqTJHwMiQKXP4rrt6MmXbofFDWihQ,31753
 sglang/srt/mem_cache/paged_allocator.py,sha256=BrJS0vN1k-vTSgb_M8u_1KoZFRgzgR1WRyImCTq3T0U,9770
 sglang/srt/mem_cache/radix_cache.py,sha256=Lm-pco6CJ4orb9IfDpbHm5MnyK8Ya0OF1x9p88dv548,14906
 sglang/srt/metrics/collector.py,sha256=zHg4twFQJvuK1mSme3-EYQa9PJryfp_u7a4RxQ5RcO0,8874
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
-sglang/srt/model_executor/cuda_graph_runner.py,sha256=ulYmFv0mlQ4aawuFpX2qkaAVx3qE5tEYj4D7hOOEct8,23325
+sglang/srt/model_executor/cuda_graph_runner.py,sha256=ZaN1yPju52iOVQwfCVju2Iw_ci4B87mI_xi21IZFZis,23325
 sglang/srt/model_executor/forward_batch_info.py,sha256=_qSMTiLxvcPIIgqRfUqG4W--OoirVY7ulcFfZqQIqjo,28689
-sglang/srt/model_executor/model_runner.py,sha256=5qvLlql8rIMJUa8DTChrraq_7-s6PusnpdPctED3PJU,46909
+sglang/srt/model_executor/model_runner.py,sha256=mDPja0tc2NNa2UBf1GfpjTtowmCdDncSS06NkwzFnmU,46476
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=YYmtvkQw0B1qgPw0_gN-K4yy7CEYbTSR__0Dl1Fnm6k,55342
 sglang/srt/model_loader/utils.py,sha256=0NaMR67fESFopaklmsleiL27XH1QUrjZW246MUu1EJ0,1369
 sglang/srt/model_loader/weight_utils.py,sha256=yKnau-wH9muczoCpDTCVIqXFqz-QJmEEySplX3bMJWk,32153
 sglang/srt/models/baichuan.py,sha256=HbvlErnkCSK4pRQYCSDxMcrn-1DQyfiNoeDcnRrJas8,15807
+sglang/srt/models/bert.py,sha256=kHlErDgNX_mIhfWWCnAcH_ncvYg22Y61gI34gW8GuUY,12738
 sglang/srt/models/chatglm.py,sha256=cajLN9caBl09e0TwOFkiTTKDqwlbmHo_yS-NCjdeQW8,13957
 sglang/srt/models/clip.py,sha256=fCMtAcaKjruSIWfD4YGb4HXh6Tzp2pjpgDmp5JpwBPU,19794
 sglang/srt/models/commandr.py,sha256=5Y_b3K0QY7D37nFGkyiGgY38RleRui_GJUYcHSuHUZo,15315
 sglang/srt/models/dbrx.py,sha256=4pn_fdoATg01VEqNnIAxNEsKV5XU7gwHyd289eydq1s,15598
-sglang/srt/models/deepseek.py,sha256=jZFQUVJ753qcI8_3sh6TlLF-8oYiQndQm-3No8FInXk,16910
+sglang/srt/models/deepseek.py,sha256=m8CjJIJiQ9B_ACPy7dwMLfm4kVLXcuW27zDk_lcQ_Dc,16920
 sglang/srt/models/deepseek_janus_pro.py,sha256=8wAzvcGdyo--3faMN4QtagT1eAZMhMFduvpCXqUS48Q,70456
-sglang/srt/models/deepseek_nextn.py,sha256=kca-2Fm2_SmqbOEFfd80pobooi1BXd1oe_4EsUM6SeI,13561
-sglang/srt/models/deepseek_v2.py,sha256=LtaJOxEn6ZnsXkiiVdS646u7lw68TeBg3iPy2LGj_cY,70572
+sglang/srt/models/deepseek_nextn.py,sha256=mL2nnblFmeBD8cSf15BmJh_M-8dyCx07sqlicw4rab4,13454
+sglang/srt/models/deepseek_v2.py,sha256=dAjHJ_2aRZSd0OQ0uCinDmRYpVs1tEWEeDfXy_NrdxQ,68369
 sglang/srt/models/deepseek_vl2.py,sha256=RVvi_3qsfrkqMCCnjjTA8OwUc5ySutc7asAH-rUJLVo,12922
 sglang/srt/models/exaone.py,sha256=rX7J0xFt9TSt6tMIhnYMkb5KDnqTJIV4BtjPLFwQ8_8,13425
 sglang/srt/models/gemma.py,sha256=4cdrPISg1VKnsuI-QPTpYvet4BrX8BMKvCIN82iLskw,12641
@@ -517,7 +516,7 @@ sglang/srt/models/granite.py,sha256=5WOJyNYAlt5RNHSexNfPNihhSxIMd7wPzju1cTixKig,
 sglang/srt/models/grok.py,sha256=vESZeGS4adI_JAerXIkCcTm15-CNiGeS7VHc36C6w1A,28033
 sglang/srt/models/internlm2.py,sha256=RDAT9drjdgVEFmCMq99RTn3weMQFhl1NHhkhyDX8f7M,13056
 sglang/srt/models/internlm2_reward.py,sha256=ndfGmyqYZbVZ7C7rJ-v9oK3wa-EpoBGybS8MlyKZi2E,2522
-sglang/srt/models/llama.py,sha256=71GmA-_-CNM2kuEJplNg6tfWbjCW31EzkeVIk5ZwNmo,24932
+sglang/srt/models/llama.py,sha256=Y4ROe8ohP84G4vin_Sr_vjG0XRoM5gGgnrojxOXn_uc,24942
 sglang/srt/models/llama4.py,sha256=JIVS5Q1lnmEpAHDI487gKO_9xfTCehSpzInNQeCg8JU,17940
 sglang/srt/models/llama_classification.py,sha256=4QWTFaUZIFKYZvEzs8bx8VkOZNIwdYCLrnwrdAw4QK0,3108
 sglang/srt/models/llama_eagle.py,sha256=OB2lKsjn7BcfCZljklnhk83me8j0PuQmYLou7baNcq4,4866
@@ -527,7 +526,7 @@ sglang/srt/models/llama_reward.py,sha256=LF2nqMV5XOrljGjAwJg43mBv3z6Q040I2EYlgZe
 sglang/srt/models/llava.py,sha256=KMwNNrlMuMaKEOZMDRBKBQbe6uctpKTLc0zOceyGC34,27242
 sglang/srt/models/llavavid.py,sha256=q0lHlRnoYHKJZsWnkIQdd6dYAQ26t7XsmrqA0zDGmZc,12829
 sglang/srt/models/minicpm.py,sha256=m5HFsSJj0Po09LY9R6qj6K4gceqWDMOePz3NDGgMGT4,14691
-sglang/srt/models/minicpm3.py,sha256=ZQpk6j2UjtVDR5gA0_jGYvl5Vsvm7NBH7xkpNjqgGw0,26348
+sglang/srt/models/minicpm3.py,sha256=R5-9CHgyUnsnxYPwLP7LQA3AKQ10Igtphs5x7u98Oa4,19367
 sglang/srt/models/minicpmo.py,sha256=kJnp8UwJTV7kXEpuVWA50ecRsuZyFedHlwkprix8tag,75619
 sglang/srt/models/minicpmv.py,sha256=79zZn3co9r7SERatx49EuHRoLWRiy6qeaUFgjDWJo2I,40571
 sglang/srt/models/mistral.py,sha256=EYifJUUzN2Z2-iL37eJiNZF_DB0H4pa0mKlgYRIxM70,838
@@ -540,13 +539,15 @@ sglang/srt/models/olmo2.py,sha256=azmljhJF4ivcQfUtfsAUxq3ducE4tRKTL6iwe0IKYMg,14
 sglang/srt/models/olmoe.py,sha256=TMzt-yB891bvA4X50xL0NjNnFYSx9imlA7N1EG8KNK0,15949
 sglang/srt/models/phi3_small.py,sha256=UbqZvpwWolXUPd0zbKgbL93yVXUY1n4kXJLgIe_gjaM,15508
 sglang/srt/models/qwen.py,sha256=xYkVmMZS2uMqWhfndc8EYm0olpKFnggfuMp_6aobVi4,10758
-sglang/srt/models/qwen2.py,sha256=fYE5fkyRYTEVVl8XnQO2-ybj4ZhNtM7Kn12AQt39EDA,16181
+sglang/srt/models/qwen2.py,sha256=ab912Yyk0aXOzI-wrxqN-sNF3bTVkNxB8P2uNcOCv9U,16394
 sglang/srt/models/qwen2_5_vl.py,sha256=uNnYhY8x-9H1GzUJkj7lUtR5d-0yMRNWUcT7-4qPlMU,22555
 sglang/srt/models/qwen2_classification.py,sha256=dGrMm4ebd30_lBhHOhaV57ig2iOTx3nqB4GEzsrRIM8,2747
 sglang/srt/models/qwen2_eagle.py,sha256=Iz0HWL2FgSD3FqoFhfYmbIZeEYkPTJ96lYbkncmHJX4,4644
-sglang/srt/models/qwen2_moe.py,sha256=GhDR7pP_G0NZ2HkaFVrBZnbqB0RxxNnH-8HMLwrweE4,18245
+sglang/srt/models/qwen2_moe.py,sha256=bmS2pyHD5zQo5plTCzAo_mjnahVtJ1jaRSURX1PlQC4,18313
 sglang/srt/models/qwen2_rm.py,sha256=-mQXDEv11p-I1HXgYLTtY6ROem6UYorO958WsDrzsgs,2837
 sglang/srt/models/qwen2_vl.py,sha256=NCG85isoPkepv5RU-eLh44rCHPhfT3bu7pifNdBEsVw,21612
+sglang/srt/models/qwen3.py,sha256=reaowGkotYAGHS5zTCWrvnyxtu92QKus19n-2amtMa4,12358
+sglang/srt/models/qwen3_moe.py,sha256=Tee7oW6Xvo2pV_Q93y-HKykBFiPjo_-YfeIsIelB3hA,15623
 sglang/srt/models/registry.py,sha256=inKh9iwOp3LFYm3nqujg-OtABClOP-ifc1stA9cZegA,3434
 sglang/srt/models/stablelm.py,sha256=0x_31uIr3WcWwecdPAI3ek9KkyKBJS7VwknTk2y0gjY,12281
 sglang/srt/models/torch_native_llama.py,sha256=5tfFSMAXB3ScToqTALtCXa8Oo-qPCJh-KQCNB6QOlNA,19293
@@ -557,7 +558,7 @@ sglang/srt/openai_api/adapter.py,sha256=DaSU4Pri70s3ZeMHeVzsnKjd8dA9lx_HOmpVs1TE
 sglang/srt/openai_api/protocol.py,sha256=Y8PFFhLbzhpoERM6-WsTkm-ZuGcE-3tfenh9e-AC1vc,13374
 sglang/srt/platforms/interface.py,sha256=hym3iooBB4C8if5hDZezgVN6h4NIOu7sg2ZUBIV6XmM,11246
 sglang/srt/sampling/custom_logit_processor.py,sha256=tDvoLgLqn-sy1qcY6vSrpbnHCeqbdk0uhMOO-uy4p4E,1099
-sglang/srt/sampling/sampling_batch_info.py,sha256=wrGGU27mWOi_yCBBCOvTQLBdyTjfkPuj7Hsk0zOFyH8,11989
+sglang/srt/sampling/sampling_batch_info.py,sha256=4LCowU2bk0TOSfIGpEy90N1SpTsiOKK8Rx1ZYcklUFQ,11988
 sglang/srt/sampling/sampling_params.py,sha256=nXm44Inn91YtrMpAm5mDb6-97owRy-Bh6lZ0BIpw73I,5919
 sglang/srt/sampling/penaltylib/__init__.py,sha256=mtN8grFEcaBUhl4yBHmw8NNirt_i6uKO2cDNLHOpZQE,496
 sglang/srt/sampling/penaltylib/frequency_penalty.py,sha256=Loc3qjJTksNc5s-DV7QZHjgqoo5pxk7-nZzxwyhD2tQ,2144
@@ -567,13 +568,13 @@ sglang/srt/sampling/penaltylib/presence_penalty.py,sha256=NRh10AJrrQlGJ6S-enGdRe
 sglang/srt/speculative/build_eagle_tree.py,sha256=SFQ3eHbhfNxOdxgqDP5wSV_ZlIVqLw7VivycNZ963N0,11690
 sglang/srt/speculative/eagle_draft_cuda_graph_runner.py,sha256=FP-Dc6K4zaL2KQA8QsNccBM8TXnwREh1I2iPL9KHo8I,9252
 sglang/srt/speculative/eagle_utils.py,sha256=mv--nBUgAbqP30pU3aGEMwQIHBwwevETUMQSZAelApE,28721
-sglang/srt/speculative/eagle_worker.py,sha256=Qvg3B40GKH6vjyrZ9SmiVyW6KbuRJJHYXbnRCvSz3aE,27016
+sglang/srt/speculative/eagle_worker.py,sha256=iQ7E6RLNyxXNs_jqbQ8SpdaXEMIC-ULBslrY9iMTv3g,26865
 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=7VLbWl4nCQs1wjtW4q-46jf9jUCycSs5Iw8v7sUSzBw,4284
 sglang/test/few_shot_gsm8k_engine.py,sha256=QQbrwOX6-cJDD3RZC_e7zPnt6aSo8JdF8X_lRHSjdDM,3886
 sglang/test/run_eval.py,sha256=9yO0hXZOcn4abEOs96T-XPguDEklK16Ltco0pGF3zCg,4020
-sglang/test/runners.py,sha256=zl_7wdwUbuCSZoA-f94VhnOI36VX_DwCt3cAEzIjm9s,30484
+sglang/test/runners.py,sha256=I2gXi0r663tvGlHLvh-W963Nv2yieA8MWVgwWe9zDbc,30656
 sglang/test/send_one.py,sha256=6FhbJ3c8RpXxvFTELRXaF97GpT7zXXsCDYZh1DqG22E,2550
 sglang/test/simple_eval_common.py,sha256=joqrGysuLnJFtzDRIgFkMsRyKUSyjVPFWp0_PHAL3Ik,12378
 sglang/test/simple_eval_gpqa.py,sha256=8Xt9Bw05c7SZTYrCZgB68OZUqUbLo69ywiyx0bTvSUk,3220
@@ -582,9 +583,9 @@ 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=GeTIJHxlLQfW3kM-X1FGa8Sa3dSGKHEXl5wEy-hfGis,1489
-sglang/test/test_block_fp8.py,sha256=6Ux1_E6EWdY184n8tiYOCwbyHVAUEqz9lMhSUDLIOC8,16292
+sglang/test/test_block_fp8.py,sha256=3gOC4Xkxh2LXfT7T2aL8acWzpSdJlRdA3KlO0I1Wtkc,21594
 sglang/test/test_block_fp8_ep.py,sha256=N1rvqbPErBaFFpeAw8TLYXGNZOoG7cfIBP2p5XbSyMo,10806
-sglang/test/test_custom_ops.py,sha256=4X3-odkJntwNtBAuKtCbYHu6peIP6LaI_VwLw7kmDx8,5550
+sglang/test/test_custom_ops.py,sha256=2bSo9P5_rJZYFq8Y8IKRimDfFyZZGJluhL7Ngny0Pf4,5571
 sglang/test/test_dynamic_grad_mode.py,sha256=L76yUCuk_ymNpXD2CmO8r2GiGjIvD_gtTsuFDs2NolI,1638
 sglang/test/test_layernorm.py,sha256=2GMWqqNDuGvSMSsEBF5eDCzwVSYA9E6hGhRo6s4ecKg,3764
 sglang/test/test_programs.py,sha256=VZ3vXtUDBnXz0M7gFdDH8hXg9Wa0j_qI8CVqjEgRN_E,18877
@@ -593,8 +594,8 @@ 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=er0i3KGHMkw-4UZB1GCFd4oYwRcXfU5wpO1ORqpNGGA,7626
-sglang-0.4.5.post1.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.5.post1.dist-info/METADATA,sha256=602BdUHYnIS2M5riEnyqkhTpEKDRbH0J0ubdy1FK8fg,25571
-sglang-0.4.5.post1.dist-info/WHEEL,sha256=CmyFI0kx5cdEMTLiONQRbGQwjIoR1aIYB7eCAQ4KPJ0,91
-sglang-0.4.5.post1.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.5.post1.dist-info/RECORD,,
+sglang-0.4.5.post2.dist-info/licenses/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.5.post2.dist-info/METADATA,sha256=C7YpNRcr_rgjjRxghmVgh_lQdH7BCAgvIZqNYLag3zU,25592
+sglang-0.4.5.post2.dist-info/WHEEL,sha256=pxyMxgL8-pra_rKaQ4drOZAegBVuX-G_4nRHjjgWbmo,91
+sglang-0.4.5.post2.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.5.post2.dist-info/RECORD,,

{sglang-0.4.5.post1.dist-info → sglang-0.4.5.post2.dist-info}/WHEEL

@@ -1,5 +1,5 @@
 Wheel-Version: 1.0
-Generator: setuptools (78.1.0)
+Generator: setuptools (79.0.0)
 Root-Is-Purelib: true
 Tag: py3-none-any