sglang 0.4.2.post4__py3-none-any.whl → 0.4.3__py3-none-any.whl

sglang/srt/model_executor/model_runner.py

@@ -67,6 +67,7 @@ from sglang.srt.utils import (
     monkey_patch_p2p_access_check,
     monkey_patch_vllm_gguf_config,
     set_cpu_offload_max_bytes,
+    set_cuda_arch,
 )
 
 logger = logging.getLogger(__name__)
@@ -110,8 +111,14 @@ class ModelRunner:
         ):
             # TODO: add MLA optimization on CPU
             if self.server_args.device != "cpu":
-                logger.info("MLA optimization is turned on. Use triton backend.")
-                self.server_args.attention_backend = "triton"
+                if server_args.enable_flashinfer_mla:
+                    logger.info(
+                        "FlashInfer MLA optimization is turned on. Use flashinfer backend for DeepseekV3ForCausalLM."
+                    )
+                    self.server_args.attention_backend = "flashinfer"
+                else:
+                    logger.info("MLA optimization is turned on. Use triton backend.")
+                    self.server_args.attention_backend = "triton"
 
         if self.server_args.enable_double_sparsity:
             logger.info(
@@ -169,6 +176,7 @@ class ModelRunner:
                 "enable_dp_attention": server_args.enable_dp_attention,
                 "enable_ep_moe": server_args.enable_ep_moe,
                 "device": server_args.device,
+                "enable_flashinfer_mla": server_args.enable_flashinfer_mla,
             }
         )
 
@@ -292,6 +300,8 @@ class ModelRunner:
                 if torch.cuda.get_device_capability()[1] < 5:
                     raise RuntimeError("SGLang only supports sm75 and above.")
 
+        set_cuda_arch()
+
         # Prepare the model config
         self.load_config = LoadConfig(
             load_format=self.server_args.load_format,

sglang/srt/models/deepseek_v2.py

@@ -255,6 +255,8 @@ class DeepseekV2Attention(nn.Module):
             self.kv_lora_rank + self.qk_rope_head_dim,
             bias=False,
             quant_config=quant_config,
+            # FIXME: quick fix for skip quantization
+            prefix=f"self_attn.kv_a_proj_with_mqa",
         )
         self.kv_a_layernorm = RMSNorm(self.kv_lora_rank, eps=config.rms_norm_eps)
         self.kv_b_proj = ColumnParallelLinear(
@@ -455,6 +457,8 @@ class DeepseekV2AttentionMLA(nn.Module):
             self.kv_lora_rank + self.qk_rope_head_dim,
             bias=False,
             quant_config=quant_config,
+            # FIXME: quick fix for skip quantization
+            prefix=f"self_attn.kv_a_proj_with_mqa",
         )
         self.kv_a_layernorm = RMSNorm(self.kv_lora_rank, eps=config.rms_norm_eps)
 
@@ -506,14 +510,20 @@ class DeepseekV2AttentionMLA(nn.Module):
         hidden_states: torch.Tensor,
         forward_batch: ForwardBatch,
     ) -> torch.Tensor:
-        # Use normal computation for prefill and use weight absorption for extend/decode
-        if (
-            forward_batch.forward_mode.is_extend()
-            and forward_batch.extend_prefix_lens.sum() == 0
-        ):
-            return self.forward_normal(positions, hidden_states, forward_batch)
+        if global_server_args_dict["enable_flashinfer_mla"]:
+            if forward_batch.forward_mode.is_extend():
+                return self.forward_normal(positions, hidden_states, forward_batch)
+            else:
+                return self.forward_absorb(positions, hidden_states, forward_batch)
         else:
-            return self.forward_absorb(positions, hidden_states, forward_batch)
+            # Triton: Use normal computation for prefill and use weight absorption for extend/decode
+            if (
+                forward_batch.forward_mode.is_extend()
+                and forward_batch.extend_prefix_lens.sum() == 0
+            ):
+                return self.forward_normal(positions, hidden_states, forward_batch)
+            else:
+                return self.forward_absorb(positions, hidden_states, forward_batch)
 
     def forward_normal(
         self,

sglang/srt/server_args.py

@@ -140,6 +140,7 @@ class ServerArgs:
     disable_jump_forward: bool = False
     disable_cuda_graph: bool = False
     disable_cuda_graph_padding: bool = False
+    enable_nccl_nvls: bool = False
     disable_outlines_disk_cache: bool = False
     disable_custom_all_reduce: bool = False
     disable_mla: bool = False
@@ -160,12 +161,15 @@ class ServerArgs:
     delete_ckpt_after_loading: bool = False
     enable_memory_saver: bool = False
     allow_auto_truncate: bool = False
+    return_hidden_states: bool = False
 
     # Custom logit processor
     enable_custom_logit_processor: bool = False
     tool_call_parser: str = None
     enable_hierarchical_cache: bool = False
 
+    enable_flashinfer_mla: bool = False
+
     def __post_init__(self):
         # Set missing default values
         if self.tokenizer_path is None:
@@ -691,6 +695,11 @@ class ServerArgs:
             default=ServerArgs.grammar_backend,
             help="Choose the backend for grammar-guided decoding.",
         )
+        parser.add_argument(
+            "--enable-flashinfer-mla",
+            action="store_true",
+            help="Enable FlashInfer MLA optimization",
+        )
 
         # Speculative decoding
         parser.add_argument(
@@ -782,6 +791,11 @@ class ServerArgs:
             action="store_true",
             help="Disable cuda graph when padding is needed. Still uses cuda graph when padding is not needed.",
         )
+        parser.add_argument(
+            "--enable-nccl-nvls",
+            action="store_true",
+            help="Enable NCCL NVLS for prefill heavy requests when available.",
+        )
         parser.add_argument(
             "--disable-outlines-disk-cache",
             action="store_true",
@@ -795,7 +809,7 @@ class ServerArgs:
         parser.add_argument(
             "--disable-mla",
             action="store_true",
-            help="Disable Multi-head Latent Attention (MLA) for DeepSeek-V2.",
+            help="Disable Multi-head Latent Attention (MLA) for DeepSeek V2/V3/R1 series models.",
         )
         parser.add_argument(
             "--disable-overlap-schedule",
@@ -896,6 +910,11 @@ class ServerArgs:
             action="store_true",
             help="Enable users to pass custom logit processors to the server (disabled by default for security)",
         )
+        parser.add_argument(
+            "--return-hidden-states",
+            action="store_true",
+            help="Return hidden states in the response.",
+        )
         # Function Calling
         parser.add_argument(
             "--tool-call-parser",

sglang/srt/speculative/eagle_worker.py

@@ -65,15 +65,31 @@ class EAGLEWorker(TpModelWorker):
         self.model_runner.server_args.disable_cuda_graph = backup_disable_cuda_graph
 
         # Create multi-step attn backends and cuda graph runners
-        from sglang.srt.layers.attention.flashinfer_backend import (
-            FlashInferMultiStepDraftBackend,
-        )
+        if server_args.attention_backend == "flashinfer":
+            from sglang.srt.layers.attention.flashinfer_backend import (
+                FlashInferMultiStepDraftBackend,
+            )
+
+            self.draft_attn_backend = FlashInferMultiStepDraftBackend(
+                self.model_runner,
+                self.topk,
+                self.speculative_num_steps,
+            )
+        elif server_args.attention_backend == "triton":
+            from sglang.srt.layers.attention.triton_backend import (
+                TritonMultiStepDraftBackend,
+            )
+
+            self.draft_attn_backend = TritonMultiStepDraftBackend(
+                self.model_runner,
+                self.topk,
+                self.speculative_num_steps,
+            )
+        else:
+            raise ValueError(
+                f"EAGLE is not supportted in attention backend {server_args.attention_backend}"
+            )
 
-        self.draft_attn_backend = FlashInferMultiStepDraftBackend(
-            self.model_runner,
-            self.topk,
-            self.speculative_num_steps,
-        )
         self.model_runner.draft_attn_backend = self.draft_attn_backend
         self.init_cuda_graphs()
 
@@ -218,6 +234,10 @@ class EAGLEWorker(TpModelWorker):
             token_list.append(tree_info[1])
             parents_list.append(tree_info[2])
 
+            # we don't need to run the last forward. we get 1 token from draft prefill and (#spec steps - 1) tokens here
+            if i == self.speculative_num_steps - 1:
+                break
+
             # Set inputs
             forward_batch.input_ids = input_ids
             forward_batch.out_cache_loc = out_cache_loc[

sglang/srt/utils.py

@@ -1444,3 +1444,10 @@ def launch_dummy_health_check_server(host, port):
         timeout_keep_alive=5,
         loop="uvloop",
     )
+
+
+def set_cuda_arch():
+    if is_flashinfer_available():
+        capability = torch.cuda.get_device_capability()
+        arch = f"{capability[0]}.{capability[1]}"
+        os.environ["TORCH_CUDA_ARCH_LIST"] = f"{arch}{'+PTX' if arch == '9.0' else ''}"

sglang/version.py

@@ -1 +1 @@
-__version__ = "0.4.2.post4"
+__version__ = "0.4.3"

{sglang-0.4.2.post4.dist-info → sglang-0.4.3.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: sglang
-Version: 0.4.2.post4
+Version: 0.4.3
 Summary: SGLang is yet another fast serving framework for large language models and vision language models.
 License:                                  Apache License
                                    Version 2.0, January 2004
@@ -236,13 +236,14 @@ Requires-Dist: torchao>=0.7.0; extra == "runtime-common"
 Requires-Dist: uvicorn; extra == "runtime-common"
 Requires-Dist: uvloop; extra == "runtime-common"
 Requires-Dist: xgrammar>=0.1.10; extra == "runtime-common"
+Requires-Dist: ninja; extra == "runtime-common"
 Provides-Extra: srt
 Requires-Dist: sglang[runtime_common]; extra == "srt"
 Requires-Dist: cuda-python; extra == "srt"
-Requires-Dist: sgl-kernel>=0.0.3.post3; extra == "srt"
+Requires-Dist: sgl-kernel>=0.0.3.post6; extra == "srt"
 Requires-Dist: torch; extra == "srt"
 Requires-Dist: vllm<=0.7.2,>=0.6.4.post1; extra == "srt"
-Requires-Dist: flashinfer_python>=0.2.0.post2; extra == "srt"
+Requires-Dist: flashinfer_python>=0.2.1.post1; extra == "srt"
 Requires-Dist: outlines<=0.1.11,>=0.0.44; extra == "srt"
 Provides-Extra: srt-hip
 Requires-Dist: sglang[runtime_common]; extra == "srt-hip"

{sglang-0.4.2.post4.dist-info → sglang-0.4.3.dist-info}/RECORD

@@ -6,11 +6,11 @@ sglang/bench_one_batch.py,sha256=d-LuRHEyDZjh180OCN5fqTjr8Zusk3zc0vhoJ33x0B0,179
 sglang/bench_one_batch_server.py,sha256=iu73SsvYwnuRktYZDz1P6psMiRx8MbEbF5sbsYJdzYg,5962
 sglang/bench_serving.py,sha256=jYU3rYIDkzpYhjSpJw_IkEs_UNQfouNW4phs3z5TObc,54303
 sglang/check_env.py,sha256=lDVA3ybt1wOE33HIMpkkU7zGRgLWez1_ifRRJ8qxbtw,8445
-sglang/global_config.py,sha256=fnT0U9vlHdGaQFKN9tYTnUF4-eVW4HYQURd5zvPtrg0,1286
+sglang/global_config.py,sha256=crt5cernXnDa1iQ8kGOq_ScTFclRlTQbJ-atFHM7I5I,1330
 sglang/launch_server.py,sha256=mDXfwha8LHpWQJekcCosR98QhCQsbmilsBlI5jAIgg0,420
 sglang/llama3_eval.py,sha256=gWSboDchIGybIce88bJlrCG0yiLZ513mw4gcutJlzGM,10017
 sglang/utils.py,sha256=7HpOrPBhMivWH719m7Dy1rjrAXOAsnqelpwNBBbvjqs,13319
-sglang/version.py,sha256=4aRfll4rDDWBVERpTNftyHYiEf_Z50L3jvLng1lO-j4,28
+sglang/version.py,sha256=Nyg0pmk5ea9-SLCAFEIF96ByFx4-TJFtrqYPN-Zn6g4,22
 sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/lang/chat_template.py,sha256=v4SyYViPHX3i3XT46F7vlARn4UaSiP3PBpTGtzO6uRY,17006
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
@@ -34,9 +34,9 @@ sglang/srt/hf_transformers_utils.py,sha256=_24uqCkZ4dvS9Uc5p2cCzX0Q8ShUzrh_Hp6mv
 sglang/srt/mm_utils.py,sha256=1ScBunw_x4W8ebM_AcJ62-1T2mfT8NlMJqdAhkF1lb0,12367
 sglang/srt/model_parallel.py,sha256=eLXZhvJ4wG6dh0FontNCIdVZvHYdWgaeY-5cu7TD9tE,6078
 sglang/srt/server.py,sha256=PrQb9r6L9syWHKlggbbiQYsKtpwSmECqozRbf8qnoV8,874
-sglang/srt/server_args.py,sha256=A36_q9RKNueSfsBz-NAIGfa4YvCFHfhIfuHNS_4NzK0,40363
+sglang/srt/server_args.py,sha256=97YAjIbttBjty9Rv9CQ-yGERprFRSu2Jq0zX8Yx-QH8,41018
 sglang/srt/torch_memory_saver_adapter.py,sha256=--FgbrcvJxTcRe856plD9ktqgrHGPTE18eZCJlE50hY,1255
-sglang/srt/utils.py,sha256=G8YDJ5FKRXiVDrsb30nulpogmbsVwTWx22YRT9T5inI,46529
+sglang/srt/utils.py,sha256=RVU-OORgeVQICMPzj17KHxbDdSYGOKFBnNR4dZejP9A,46780
 sglang/srt/configs/__init__.py,sha256=Nvwtif0X9IYUtj0aL9XvAo_RRZcxTshsaliwc8djooU,347
 sglang/srt/configs/chatglm.py,sha256=j-b0YkdYUmQm2y1kNmMJtKeACxWKmBbvNNkDWbs6kbI,2907
 sglang/srt/configs/dbrx.py,sha256=tdhIkXAQl1yr0MxqFmsDG1E0e2puRTTKm6UTyANBLac,11005
@@ -61,12 +61,12 @@ sglang/srt/distributed/device_communicators/pynccl.py,sha256=G-Dut_QJHOUG0j7--Zq
 sglang/srt/distributed/device_communicators/pynccl_wrapper.py,sha256=LblisImY9d6EMz-oPS9J16WHo2Q_SRL1DtlJKK63Hfg,15349
 sglang/srt/distributed/device_communicators/shm_broadcast.py,sha256=bbruDIM1GgKIdB6gi71_I0mpB179I-qyvwKuSj1Kaic,20816
 sglang/srt/distributed/device_communicators/xpu_communicator.py,sha256=ajW6132BvA6jkeipEIgN27TFycI0U06Ih2Z8WNjlA4s,1593
-sglang/srt/entrypoints/engine.py,sha256=hYLhAJpWMrz_8RuXuYIhFUBbYp4JFEQn8YwQDcnFz-k,16853
+sglang/srt/entrypoints/engine.py,sha256=3JL7aj0K5UHCBsqCOG2MTLAv-8IpwtN2K-3_xB6fNqk,16888
 sglang/srt/entrypoints/http_server.py,sha256=TJlekPuw01_AvfAhDUdD-DaxCmmW_uH_rWL2CNv2OGE,19545
 sglang/srt/layers/activation.py,sha256=f9KGwGi2znUx5SFKH_vO8htpBkfQ550VZZIycFDfPlk,5602
 sglang/srt/layers/dp_attention.py,sha256=LLUMHIdphhQy1rNR52uwIFl85oDFPAsogMwYF3d83PU,1910
 sglang/srt/layers/layernorm.py,sha256=2_9XCR2l18c3jMrbTgfQ350C1gc9Ua_z965KRxJHJeQ,3858
-sglang/srt/layers/linear.py,sha256=u9ND-ZRv_uq8I3bxqbN11bus-kr5DHqdo67ksFA0EcU,50724
+sglang/srt/layers/linear.py,sha256=gtDbHu7ewk7Gta9MECGMT-vtldlQkyY3hsv2MVpYN4c,51000
 sglang/srt/layers/logits_processor.py,sha256=_3TZNUbvjmw63ywBv6V6WU87G1TErMaXGa7-VNExM1E,12190
 sglang/srt/layers/parameter.py,sha256=sX6aB69qbD6jRqQeOfXqK_ueyyZlXCeC0AlglbsRPcM,14901
 sglang/srt/layers/pooler.py,sha256=rj2lygvleBnyLCBZ8I11HGMgpfIDsT0l3PIkshJwdu4,1606
@@ -77,22 +77,22 @@ sglang/srt/layers/torchao_utils.py,sha256=Ws24FdRBSkTpyeyA6bQrdDm-W5wfDxKvSIPUSa
 sglang/srt/layers/vocab_parallel_embedding.py,sha256=txcjkuSDa6gZwESKj8X-HSLhAnMmDXL0FmFWY9SKqik,22155
 sglang/srt/layers/attention/__init__.py,sha256=KlQ0fl-o9v_NxBDhNZ4dPW2uQ2HeJjLm-0MTMWgaa28,2980
 sglang/srt/layers/attention/double_sparsity_backend.py,sha256=4mVyFPfZxPTwkQHGNCfI_4hQ8CbsWXJfxz-IQW77gAc,9143
-sglang/srt/layers/attention/flashinfer_backend.py,sha256=BlirMAu8dDicGdpZiW51hx9FpiU3EY-MNhZ1vdCJxGo,42398
+sglang/srt/layers/attention/flashinfer_backend.py,sha256=8bEnIXFXBB8lv0ruH_paqT8uLIHsKYKYhQ9ytzDcsYI,47681
 sglang/srt/layers/attention/torch_native_backend.py,sha256=KrcAqTLVZLtwgOmB0xhwUUsX32M-5LYZpNxaRNT4VuA,9252
-sglang/srt/layers/attention/triton_backend.py,sha256=mtLs768rhtCF_BVAV_rmYac0U4R1_HHc-9ic4JratsY,10100
+sglang/srt/layers/attention/triton_backend.py,sha256=mbYaYKHYrUyL2zEXrPtDRIcvVNe6L-bmcdLhKF92V-0,21292
 sglang/srt/layers/attention/vision.py,sha256=zLjKmzUlkgq1RFcP3b4EPArOAKovoaDLgYfM5SyB2wM,13181
 sglang/srt/layers/attention/triton_ops/decode_attention.py,sha256=tcUAdacBWTpZmro7vZeRPasfwRWFlCR4bxfGpFOYgZ8,17831
 sglang/srt/layers/attention/triton_ops/double_sparsity_attention.py,sha256=ztLWKeW-260EiIw3kCAbtUTUHHxAICz2mVxZJFes4oI,31167
-sglang/srt/layers/attention/triton_ops/extend_attention.py,sha256=R6QgrcBf6XuLzQ1jamrILNypaPi3ynkMPTfjae0d3JA,12695
+sglang/srt/layers/attention/triton_ops/extend_attention.py,sha256=U46NHqzRJDKAEfGIgsqrh6TSp8vu1GGEA2LPMwcwsVw,12691
 sglang/srt/layers/attention/triton_ops/prefill_attention.py,sha256=Y66gZ37u0GKMPtI8n5MbO6uOxRuGEmKIG0IPbJTOqAM,6213
 sglang/srt/layers/moe/fused_moe_native.py,sha256=OEWpM93X5tJG4-rwz5qmdpTzEUR73zun29YRV3bZglY,4269
-sglang/srt/layers/moe/topk.py,sha256=6A4W1ztlV2dQvkXcPJvFvAg0QEhE58Q7eE7iw8N36J4,7230
+sglang/srt/layers/moe/topk.py,sha256=uegUdoIb9OVi3_CllGN350DlVYuDJt1kjQZKxNdwKgA,7244
 sglang/srt/layers/moe/ep_moe/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/srt/layers/moe/ep_moe/kernels.py,sha256=wb_S2qLxoWWgQu9coXy0XLNGvHzdZSdwXr0PGy4QySg,10940
 sglang/srt/layers/moe/ep_moe/layer.py,sha256=aS8t1XUvlTnO9IQaxGjW5bOXP4FrJDXzymEIvlIDMro,22603
 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=NckXo7tQ0rp3trdAq5TB5mP4MmCdjm6RIN157d9Efvk,37720
-sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=-49WRpq9OtRZocQjW-YNcB_ruK09nIJqGHKNa8CJsws,22691
+sglang/srt/layers/moe/fused_moe_triton/fused_moe.py,sha256=Hyd7dDTaoC5SLdNDH9n6ot3Ibpv7zva2ZG2pRkxBRdc,37964
+sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=RWBo3j5AzZls5eD2eaejetSfMz1yQg2_Tmv_i59Ml6Q,22836
 "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
@@ -124,8 +124,9 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=-49WRpq9OtRZocQjW-YNcB_ru
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=int8_w8a16.json",sha256=kklgf2qLI5CQYiJJ5e9Gxx2gAfGxcyMDYpdJnIXPV8E,2748
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=16,N=800,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=8e0tN_DHPwvh_HECVHx9oOF_4WWdaht4s6Nmd_K-aBU,2904
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0aSYzpv_cBAlpWCPrfGgNTCfae1KdKQnT56E8XFQl7A,3262
-"sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=yN6aItfVswTn6ggd1NJb7MmO9evCet5cRIiKeHcb2r8,3733
-"sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=yN6aItfVswTn6ggd1NJb7MmO9evCet5cRIiKeHcb2r8,3733
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=5ro9O8Nf-7MB8NGCQ3QV5kB2k3iSWe0rdTz2A4W72CA,3732
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=5ro9O8Nf-7MB8NGCQ3QV5kB2k3iSWe0rdTz2A4W72CA,3732
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=5ro9O8Nf-7MB8NGCQ3QV5kB2k3iSWe0rdTz2A4W72CA,3732
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=IMTKnPRjhSptf7smIkpqmMjSML9SQ7I8CpkbR3Inzqk,3258
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=256,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=R4B2n2vGt4pPo6jS4Bmnx8AYtcfF9qQJE5bD7OhmXHs,3265
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=1280,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=6QPLhZurIqcMVdy3w0Dd7gLViKxsyJRBz-qd8agpi6Q,3248
@@ -149,11 +150,13 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=-49WRpq9OtRZocQjW-YNcB_ru
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=640,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=eD3Y9sOwHFcgVdOfya8KxPhvLx_b4whfEWm4d8Y2HW8,3268
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=64,N=640,device_name=NVIDIA_H200.json",sha256=KIfpZydSl31FOEqq0EBfxTyWRj1QTDwTjkPHFjNO3_A,3253
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=AMD_Instinct_MI300X.json",sha256=OnadAdmDbX17Ni9VPrNXYSsxYhbtBeniCxxhhb0UmUk,4733
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=AMD_Instinct_MI325X.json",sha256=OnadAdmDbX17Ni9VPrNXYSsxYhbtBeniCxxhhb0UmUk,4733
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=AMD_Radeon_Graphics.json",sha256=OnadAdmDbX17Ni9VPrNXYSsxYhbtBeniCxxhhb0UmUk,4733
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=V_sgDtEtGEuBsGVa0maYJHhhGqe1NE7l-1ek2ed9WP8,3082
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=LD4Z5MRR5Ivi4bYB5hMgymtvmFyVJwq6gmehA7fzecc,3271
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=14336,device_name=NVIDIA_H200.json",sha256=GLIH4egg-pE-NWU5XqKuJCoRXciHN6GSc3NaE4PaeYg,3261
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=AMD_Instinct_MI300X.json",sha256=bKsYVJm-IvWHWpxUG-lMPkyNz0nQpDb4UEIv895c9JI,4730
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=AMD_Instinct_MI325X.json",sha256=bKsYVJm-IvWHWpxUG-lMPkyNz0nQpDb4UEIv895c9JI,4730
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=AMD_Radeon_Graphics.json",sha256=bKsYVJm-IvWHWpxUG-lMPkyNz0nQpDb4UEIv895c9JI,4730
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-40GB.json",sha256=AffDc0_51ML8HiA3757zbD10TZJdUsUDIYIqO4g0yUw,3250
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=1792,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=IEYBNjt9HGnzoOVSWvL0A0jUqq926QD0_BvVYR4RA1Y,3252
@@ -166,6 +169,7 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=-49WRpq9OtRZocQjW-YNcB_ru
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=4O4VeMpgFNrqWyWqWgYgcYAgBQnOlAXvt26CRSXK-sY,3270
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=2048,device_name=NVIDIA_H200.json",sha256=qfjbXqbl902TuiyzzomUy2sMvs-Dud8ZphDRY5WIPBM,3260
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=AMD_Instinct_MI300X.json",sha256=_bw1_oads8tz51i4RVQUAjNi8r3b2Q2jPbi50TLFzlY,4732
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=AMD_Instinct_MI325X.json",sha256=_bw1_oads8tz51i4RVQUAjNi8r3b2Q2jPbi50TLFzlY,4732
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=AMD_Radeon_Graphics.json",sha256=_bw1_oads8tz51i4RVQUAjNi8r3b2Q2jPbi50TLFzlY,4732
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-40GB.json",sha256=Ru460ZgnUP4U8OsJfwF8n-AI-gfcolNR3_qzoxG6DtY,3254
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=K6BGrKw_oHTAtHjsZldcjp-BUM1dIecKXrrRn9OpRGs,3254
@@ -176,6 +180,7 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=-49WRpq9OtRZocQjW-YNcB_ru
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=NVIDIA_H200.json",sha256=Pi2coJlJlpgqXiPRd77B_eCmmi7sCdBuoSGK1RA5YO8,3258
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=3584,device_name=NVIDIA_L40S.json",sha256=p2qlRhTt7owWB8keEmoCrPZpo39IAxsKnULFQ7R38SI,3873
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json",sha256=AKIX43JVc26ERb862pNOMEfGhsgyk1OGa42EptAfG1s,4409
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json",sha256=AKIX43JVc26ERb862pNOMEfGhsgyk1OGa42EptAfG1s,4409
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json",sha256=AKIX43JVc26ERb862pNOMEfGhsgyk1OGa42EptAfG1s,4409
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=DxYu8regZOSFu8ugFGA_QbwWK4g8xwQUZF9a_nNY4Cs,3255
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=obzfE_9XgsbFNfC9biYOHxR-V_Bgc7PKT8qZZJaiJJc,3262
@@ -183,6 +188,7 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=-49WRpq9OtRZocQjW-YNcB_ru
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=rR8b-OuQ3watb8b2zuNlxKDSZpzlAagm9nb-FdKkt7s,3270
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=4096,device_name=NVIDIA_H200.json",sha256=8G_QqV_DhvZ6xSavMSpeE6qcXPVpsVjEtJabydybKqY,3263
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=AMD_Instinct_MI300X.json",sha256=54KpHTMGt_zDQHqbdopuVHPpiI44ZsN_5LBUBZ_woY4,4733
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=AMD_Instinct_MI325X.json",sha256=54KpHTMGt_zDQHqbdopuVHPpiI44ZsN_5LBUBZ_woY4,4733
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=AMD_Radeon_Graphics.json",sha256=54KpHTMGt_zDQHqbdopuVHPpiI44ZsN_5LBUBZ_woY4,4733
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=NVIDIA_A100-SXM4-80GB.json",sha256=BAJnXTZoewwCtzJLUPJ0oYuALv640MvDuLseGcsYaaw,3252
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8.json",sha256=-Tj7ImS6ZFDof_0VTyq7kVm8XD9B54RD6CUOPSf3Jjg,3265
@@ -190,21 +196,23 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=-49WRpq9OtRZocQjW-YNcB_ru
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8.json",sha256=3YQakSmUKhpw1KO7Hn-tEc-yyD1fEj01_6JlSYnrrlI,3274
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=7168,device_name=NVIDIA_H200.json",sha256=W2ka_U8pzwjzX62NEGKXR32uuSR_zfHD1XjXYf5bgBs,3262
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8.json",sha256=aMP7oZmh8BZnPOrl0MFibcdhTn3VmOSjqoKoK2rMSbU,4323
+"sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8.json",sha256=aMP7oZmh8BZnPOrl0MFibcdhTn3VmOSjqoKoK2rMSbU,4323
 "sglang/srt/layers/moe/fused_moe_triton/configs/E=8,N=8192,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8.json",sha256=aMP7oZmh8BZnPOrl0MFibcdhTn3VmOSjqoKoK2rMSbU,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=_Sba1KQnmZNKGDKM1MfBs2T3uDqOHfeW6IHO2mTUvfs,4471
+sglang/srt/layers/quantization/__init__.py,sha256=MU8FV-uMl7XVrECUmyNFxAOobGza05I1FlDpvfoHP1o,5848
 sglang/srt/layers/quantization/base_config.py,sha256=daK9p0aijMszLUm1W4Pc33FK87MdqYK1NoWFKif-j80,4599
 sglang/srt/layers/quantization/fp8.py,sha256=ibttPVCUsCQ0LXy7FUb8wnzqGcGZQXQLqwCB4a2fai4,35160
-sglang/srt/layers/quantization/fp8_kernel.py,sha256=qlXXT9WO3TKxZv9r0rAdcDjO_jZYobtKnvTxmHDrfoo,16304
-sglang/srt/layers/quantization/fp8_utils.py,sha256=7v-RNwuYXa-gPO3msRDB0Z3uajOQMYd2Cj0NMoq1hg4,4148
+sglang/srt/layers/quantization/fp8_kernel.py,sha256=rbuoOhgpA8_sWE5Tm3C9m0YmLqUSSBKKunLiAnHhh6c,19300
+sglang/srt/layers/quantization/fp8_utils.py,sha256=_1uEpKdwq-GSKRMdSWpK-0z9koNctHfnuQr9wBIKRfw,5211
 sglang/srt/layers/quantization/int8_kernel.py,sha256=t_BLVf8XjOyn7S3Lu3B4hXvw8DvTg4Anco7TNadL58U,1436
 sglang/srt/layers/quantization/modelopt_quant.py,sha256=_VdVz77dTP-IczPeFrdH6Ttro2D26BZvMlZkCKWj_5o,6200
 sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id3CwlNlMU8GIuZc,3344
 "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
 "sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=CPo1WRF0HgsQMPBkvpoImElQMrfwpJLhEvL86e6fkPU,3247
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xqJNygFgPCe4jDpkfVOWCXpuNMUjmssdD8oGhp2iXv8,3726
-"sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xqJNygFgPCe4jDpkfVOWCXpuNMUjmssdD8oGhp2iXv8,3726
+"sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xqJNygFgPCe4jDpkfVOWCXpuNMUjmssdD8oGhp2iXv8,3726
+"sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9ya9f1Nt0g0RD-6sRRBZOaIPLSpSFZCz7jNvqTPrgFE,3732
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Qoj9rLLRDbKM4IKBCXvN8RcxzSmNPd0TQUiM7CXDqHI,3241
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=UZljnxxCSjwnZlX3OgKWZJGXCf5BWF_agEpNX8I4Zxc,3248
 "sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=4D3Ku4y7BCVEJzueKvQC_KvOR026w3ONWsxfsA_YrEc,3249
@@ -221,7 +229,8 @@ sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id
 "sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=GsLoYkaZ2p4Qu0Coj-X90s7JWyfZBOloIHPlyNKSIes,3246
 "sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=BpKweY4HfBx5xvqSvwNahy9x7R5FH-YK8j6rLFQdKwo,3248
 "sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=I6a5wQ0R3cBp8ix_PDamWZN2aJmV_1p0tg430L3Updg,3727
-"sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=I6a5wQ0R3cBp8ix_PDamWZN2aJmV_1p0tg430L3Updg,3727
+"sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=I6a5wQ0R3cBp8ix_PDamWZN2aJmV_1p0tg430L3Updg,3727
+"sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=bSxTaptdcgj27mQGmdUmQtYTn4V_8EcmtRaVNigKjLA,3730
 "sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=8zuJhFdd6aXREpiqPFhIKEFWA5lgLVGrG0-a9UXcBqk,3262
 "sglang/srt/layers/quantization/configs/N=3072,K=1536,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Zn1TvhAoPOv0zQBYHOZhwdDw3oqyxm0zIa7IJkTCHpo,3247
 "sglang/srt/layers/quantization/configs/N=3072,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=yqjO7zML7EseBJw6Bn5MTyHeAitkPsl1dndXeL6Rn6A,3257
@@ -234,21 +243,25 @@ sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id
 "sglang/srt/layers/quantization/configs/N=36864,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=laYeH4w0iZOj2Yg3vDgtKoroNQnwBEX4GUGLrO9095I,3260
 "sglang/srt/layers/quantization/configs/N=36864,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=TWcPDZ2miQMD6OWDC1FteRs80ND9RC-oJL3PLVmJbtI,3257
 "sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=6HvsrapGzLyCAZe_NE0VdcJTSJv1aztD1ZqFG7VODUA,3729
-"sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=6HvsrapGzLyCAZe_NE0VdcJTSJv1aztD1ZqFG7VODUA,3729
+"sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=6HvsrapGzLyCAZe_NE0VdcJTSJv1aztD1ZqFG7VODUA,3729
+"sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xxNfGtHAlxDVX7PBnqExJN0UnYlA0UbaYoXUmuX0JsI,3739
 "sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=6Z7kIa14RjVq3ek_C15q5mUu1IrY2r0OP8S-_pm-MYU,3252
 "sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=r63SZkUJJV87B00hAX074_uaC7wwQXdurlJsB1jUA0I,3254
 "sglang/srt/layers/quantization/configs/N=4096,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xBhxdCFf3waTUsLxJxA54R90zODbC_DKI3XXBVKjKRw,3252
 "sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=XGNbUYyeRnb5NyfpTc1lueHjW_i49O9j9MA-MorasdI,3726
-"sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=XGNbUYyeRnb5NyfpTc1lueHjW_i49O9j9MA-MorasdI,3726
+"sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=XGNbUYyeRnb5NyfpTc1lueHjW_i49O9j9MA-MorasdI,3726
+"sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=16Qk4BrbVQOdR9Et2T3SnLphQdvEwuuG3W3XCmAFa3s,3734
 "sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=O_SV2vo_oaABfT6Mxqcmo12pnhKtfX4TnXfe02OcHJk,3254
 "sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=g12Xkurat7oUS7LdS9pHLKFlur4_FaMGiGBvdq-iBCs,3242
 "sglang/srt/layers/quantization/configs/N=4608,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=tFdrY5nADmXUlShdN8w8Jzkxuj_RPLXCRceX9FhQ35E,3251
 "sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=YA4P3iWQcyvx9wRgvs5zOqj3MKb0i3lDTfX3iTzJh2c,3723
-"sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=YA4P3iWQcyvx9wRgvs5zOqj3MKb0i3lDTfX3iTzJh2c,3723
+"sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=YA4P3iWQcyvx9wRgvs5zOqj3MKb0i3lDTfX3iTzJh2c,3723
+"sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=m0E9TwfZjvtopYFhI92VSaqhiUZpjBu69kv56rKMEuQ,3729
 "sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=TO2qRGmp37v53Zqu8Joeq_BSbtwM_mpVoozGyoNg0-o,3254
 "sglang/srt/layers/quantization/configs/N=512,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0xquf00fgfrDODpaxyre0VDcjqfzqExj939rzeJ8pMo,3244
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xsFMrq4aybClfJyhm78c1Hf1jcyFSGnfygdHYp7OhSQ,3727
-"sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xsFMrq4aybClfJyhm78c1Hf1jcyFSGnfygdHYp7OhSQ,3727
+"sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=xsFMrq4aybClfJyhm78c1Hf1jcyFSGnfygdHYp7OhSQ,3727
+"sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=A4uzWJTNhyNVh7ntOvUpT0TheaEVu_js0NCNdav8mTs,3730
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=qG6v3n3qF6LE2DdGT-mDIXecZ1a7vg7p3QqXYCMX85k,3254
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=EgFTGyW_YuDwyEDUCoGglyI1ETdj9J7AR0UfJ86jMoI,3249
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=BpKweY4HfBx5xvqSvwNahy9x7R5FH-YK8j6rLFQdKwo,3248
@@ -270,17 +283,20 @@ sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=TdWuE2RIsIyr4Im24MuWK3XyiNtbhO_hAiAXDz5gNUk,3246
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=x476nFeltB_2iO9_6y-z2P_unAbh7ghLPFi5z2LOTOo,3253
 "sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=sVbH4YRLTxBqvTh_6xbtXkj3orOrKytlwM-_4gtD6IY,3725
-"sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=sVbH4YRLTxBqvTh_6xbtXkj3orOrKytlwM-_4gtD6IY,3725
+"sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=sVbH4YRLTxBqvTh_6xbtXkj3orOrKytlwM-_4gtD6IY,3725
+"sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=sTvaJ0RiCaQem4F1z7oES6RVRJ2gKgBuccX13S1SqGc,3733
 "sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=4uWiQMh3cZY_EtLA0a3PU8Z1VCunF2PpolTPYeP9Rjo,3256
 "sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=D0moiKqS73oril32iNj5gRJUWpT2SZ5jf-ZesUZnNv4,3254
 "sglang/srt/layers/quantization/configs/N=7168,K=2048,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=N37dUL_J2JVpgLFlnlz__Ck7Z4njROnNAO8V2oiDqr8,3253
 "sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=zuSYO0DejuHJK0dqSszTySoZUFizgjtLIXSjjOC_lpc,3726
-"sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=zuSYO0DejuHJK0dqSszTySoZUFizgjtLIXSjjOC_lpc,3726
+"sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=zuSYO0DejuHJK0dqSszTySoZUFizgjtLIXSjjOC_lpc,3726
+"sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Rq-eMMK1deUJzmHQOH0E_pwQP7l-ZU-ECTP7Xwegavw,3736
 "sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=LdtOyXsA9r18GiFkmDOkiRinsDSZBZ8NYapL59EZ4iM,3264
 "sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=07GarBHmiiYkyqn-qxEtrAcgCETuUbqm6HqlbH9yJi8,3252
 "sglang/srt/layers/quantization/configs/N=7168,K=2304,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=kEuvCsW3YNByF-DALYqPZpW3TL8ZbtQ5gUNq7-8YvZ4,3252
 "sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI300X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=aoq4792zPo87QO7VrEf9fb_vj0zPiHIu7Ho9aMXwcLw,3731
-"sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=aoq4792zPo87QO7VrEf9fb_vj0zPiHIu7Ho9aMXwcLw,3731
+"sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=aoq4792zPo87QO7VrEf9fb_vj0zPiHIu7Ho9aMXwcLw,3731
+"sglang/srt/layers/quantization/configs/N=7168,K=256,device_name=AMD_Radeon_Graphics,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=_RHvFcbtpsZBKxZte2-E3SUHtL1pwRtqwhSV4BMcyKo,3734
 "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_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FFBjSWlpKXMxfAUUYUqXbOK_Hd7qBeBsfbcaa9uB4qY,3249
 sglang/srt/lora/layers.py,sha256=r34oprzwyE3SWPvaNkBvXWPtfa-0IY987_bjj36ySfw,9996
@@ -301,16 +317,16 @@ sglang/srt/lora/triton_ops/sgemm_lora_b.py,sha256=Ai5vPriT4OgACwK7xrpGgf5L1oaN9x
 sglang/srt/managers/cache_controller.py,sha256=DXnIunJgtTws1WF2vZOYVQe56vacV7Mn4wL9zoG8Xz8,10909
 sglang/srt/managers/configure_logging.py,sha256=aY9xExurz7t_IdItd-9GuVuM7kEGB8_bRryhZxKdu9o,1542
 sglang/srt/managers/data_parallel_controller.py,sha256=b64aC6iLr5RolJyNQnT-yTQ_TSI9DDLtuABf_TPTUrM,9421
-sglang/srt/managers/detokenizer_manager.py,sha256=A-tZi9VPkrIAVteQItYUY-07V1rWmySFHNcVf8qAdPI,9578
+sglang/srt/managers/detokenizer_manager.py,sha256=XC2INyykOgwmIrFEGc-zf6LGZ5mMt6oPZt1YRXW_cbY,9650
 sglang/srt/managers/image_processor.py,sha256=s1QH9cSzT_nnitc6idzFjuGDp-pDnMTpbVZoQfzdSXU,20671
-sglang/srt/managers/io_struct.py,sha256=1Z6MCVI1LN2lS_7e8WHkpVNT_LW62mE-jpZ2Jn_FAtE,18267
-sglang/srt/managers/schedule_batch.py,sha256=oP6ygJUOmo6PuXcA_wecRvOOa_WdpwmIyCPSgJy4qAc,48743
+sglang/srt/managers/io_struct.py,sha256=9jhu794cc_BljFmVL6kQseTHGZNwEzONdlGEy_wjAcA,18357
+sglang/srt/managers/schedule_batch.py,sha256=smqDrzohvA8j76CLgI53CvpduheW1m__26S0O8HcCf0,49187
 sglang/srt/managers/schedule_policy.py,sha256=Qero_lwPEb7bM87qjWtYijGyRhtY0mMwjWP6SbjvaUE,18260
-sglang/srt/managers/scheduler.py,sha256=akwBfBcNgpCXY1vp3FlD5-bOUMKfUBR5AC3XzSBRDYQ,70757
+sglang/srt/managers/scheduler.py,sha256=w0FPjiU5MoyP58UdJoPBr-hf-WmlWPpqb-5TSJDJBLo,71908
 sglang/srt/managers/session_controller.py,sha256=WXRbtninVEVM0rQYiXFzOwsDph0TNj1L2sRCWQF0dSg,5571
-sglang/srt/managers/tokenizer_manager.py,sha256=TjhX0IeFCmk31PDmtVV7Ilc8rqI361XUf_p2KO3ai7s,38669
+sglang/srt/managers/tokenizer_manager.py,sha256=q2FhMcjX9PYTWBz6E98AsE8INlF5I_YFfTWzkBWBjdk,38900
 sglang/srt/managers/tp_worker.py,sha256=OiHpFR9Hy1GpgLEkTDsykBiFuv1VKmkjQS58gQVPQIs,8126
-sglang/srt/managers/tp_worker_overlap_thread.py,sha256=7p6zREndc4a9fmYfqW4iY9IYANxdoAioaf0hU92-8Ow,8893
+sglang/srt/managers/tp_worker_overlap_thread.py,sha256=e0Fz0Yrwl98azg_6vwLDE6b_5WMcb5llPIbxLfoYwAc,9090
 sglang/srt/managers/utils.py,sha256=5i75uLlQOF_5CaT02CrWtwozMTtwTg2_nLP8Dtr-JZQ,1536
 sglang/srt/mem_cache/base_prefix_cache.py,sha256=qxgpSHm3qtMdab4U35Mr2BE9TQNjElrnrNMTwL_Osdo,1049
 sglang/srt/mem_cache/chunk_cache.py,sha256=hc_reKKvoI4r8xkgf4I4eIkwXWTJC2ZXaQWuODQZnx0,2572
@@ -319,9 +335,9 @@ sglang/srt/mem_cache/memory_pool.py,sha256=9ud97u1cXnN6O0qlR8tv8woN_20gqisTV6aBg
 sglang/srt/mem_cache/radix_cache.py,sha256=hVILXvc5PauHuLTeyZbm3NCf3AOimaAuXjll53MSLeU,11754
 sglang/srt/metrics/collector.py,sha256=_yl0_paSARxS1ypZgd-pLJ29tMizolHuwROX21dOXTk,7326
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
-sglang/srt/model_executor/cuda_graph_runner.py,sha256=7MLskhry5KFP-lhqUfAwZ8P4HckFCEQonYqIDjfAWSw,18455
+sglang/srt/model_executor/cuda_graph_runner.py,sha256=hH646E_c4UlclGEawPDjg4KHgTUEk70WrPl6C7nnltM,18774
 sglang/srt/model_executor/forward_batch_info.py,sha256=t1RlBgoeS-_Ikl28Xjvt-aouh1nNUc3eLM4iGY4_QqY,14988
-sglang/srt/model_executor/model_runner.py,sha256=5ET8CXwSrzIKb_rDW-6S0p7tv-iCPoEwZk3dESwPdF8,33090
+sglang/srt/model_executor/model_runner.py,sha256=uohQ2n2R1HcVyaHwbdwM6xDvFxZSLgxacjMSrrogLpw,33537
 sglang/srt/model_loader/__init__.py,sha256=zGZkOBz1zx-pkaIy47BasL3fjDlAcxAXUTjInOhXHAE,919
 sglang/srt/model_loader/loader.py,sha256=2d9fJNxC3Y6YWmQX4nVOB-b9Glc43ztlkJYJFX1_kxk,46811
 sglang/srt/model_loader/utils.py,sha256=0NaMR67fESFopaklmsleiL27XH1QUrjZW246MUu1EJ0,1369
@@ -331,7 +347,7 @@ sglang/srt/models/chatglm.py,sha256=n8uZpx9iHw6V-XCns9mtTf99Iqh35ZjPC5bFDYtkoes,
 sglang/srt/models/commandr.py,sha256=y8DFUW0NKbkoY2DP6nhgJ1f7F_ysjaHEkEnZYZW2zdk,14523
 sglang/srt/models/dbrx.py,sha256=-L9QkUr_xuMuI6mn0AzG_VE1MqRXoaaFtD4r8UuAzkY,14789
 sglang/srt/models/deepseek.py,sha256=KfcQ54BqlS73XQmtcG0sfnmm3VXOGwUIkd34WS6Gp0Y,15694
-sglang/srt/models/deepseek_v2.py,sha256=1JfWeo0q0N7q9nRuz3pfM4QWTTGu3e1qk2957X1x328,38499
+sglang/srt/models/deepseek_v2.py,sha256=9_284fDT15WEmv8qfnH2EzKX8fxSZrNyiz4iQtgb0tI,39065
 sglang/srt/models/exaone.py,sha256=Wvr6XofnH2feJ-TzAm5aD1YTyfcum6JdnKMG1S7Xy4g,13035
 sglang/srt/models/gemma.py,sha256=4Jvt9F-BNhPFiBi5H8aPqcYqKeJLI9KZKy2WpR96RpM,12123
 sglang/srt/models/gemma2.py,sha256=cyQfby-kp2OZPsUACmBh3-jsXkYwQg9Tj6xqtZ7mTwM,15947
@@ -385,7 +401,7 @@ sglang/srt/sampling/penaltylib/penalizers/repetition_penalty.py,sha256=l1DyU8kC8
 sglang/srt/speculative/build_eagle_tree.py,sha256=zWthboIgzPzSOXcGxDpDv0rBOQP55HYGrBKGqm2gWF0,20732
 sglang/srt/speculative/eagle_draft_cuda_graph_runner.py,sha256=FY4hcwd0Blx7AXbeX6quaXPNgWA8WGIqVcQiEgHyERk,8002
 sglang/srt/speculative/eagle_utils.py,sha256=ypjVmVTVzCGclOVHRMJxdLUSPkf1-7bNXQS0oP6dn5U,25644
-sglang/srt/speculative/eagle_worker.py,sha256=yDCU4liKU_jgtwQ-8L9_GuUtFRo3l1Df7mCeKX8yHCo,12085
+sglang/srt/speculative/eagle_worker.py,sha256=33zC6txEsNp9hD48iy-_67ov83Pf4iASulLg8GHLy5U,12898
 sglang/srt/speculative/spec_info.py,sha256=D7A27UU1iOwIBEjXTgAxZ7jdftbTiVlMCvK8GmYr2zg,488
 sglang/test/few_shot_gsm8k.py,sha256=7yDbEQe49gZeJhz2wFFX-gf_59ThDKsCS1xwfogNc7k,4034
 sglang/test/few_shot_gsm8k_engine.py,sha256=QQbrwOX6-cJDD3RZC_e7zPnt6aSo8JdF8X_lRHSjdDM,3886
@@ -403,8 +419,8 @@ sglang/test/test_layernorm.py,sha256=IacByD5d-stXjzBz8Ypamc7povlcedpKPbb_4JLgo3c
 sglang/test/test_programs.py,sha256=aUV9Ex_B714ph7ytv6W3J7sdGDKC6lGIhUy95Yg6AHQ,18878
 sglang/test/test_utils.py,sha256=BU6lAX3bu3TNQZqVC9UPnyq3I7iV5kigHQKJx7UNlOQ,26192
 sglang/test/srt/sampling/penaltylib/utils.py,sha256=CjxHgywh0hx_87iynzQt_ztHu6zBVuE-YrZ-XPmW6U4,12906
-sglang-0.4.2.post4.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.2.post4.dist-info/METADATA,sha256=VylN7wgaeE_8T-ddwxTmujJo04JCRpaP_uyhGfo-InQ,23773
-sglang-0.4.2.post4.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-sglang-0.4.2.post4.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.2.post4.dist-info/RECORD,,
+sglang-0.4.3.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.3.dist-info/METADATA,sha256=cpA0ecZd4jfaThOrStEpOGbXDTorUxqYdU4catzo2t4,23815
+sglang-0.4.3.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+sglang-0.4.3.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.3.dist-info/RECORD,,