PyPI - sglang - Versions diffs - 0.4.3.post1__py3-none-any.whl → 0.4.3.post2__py3-none-any.whl - Mend

sglang 0.4.3.post1py3-none-any.whl → 0.4.3.post2py3-none-any.whl

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.

Files changed (25) hide show

sglang/srt/entrypoints/engine.py CHANGED Viewed

@@ -330,7 +330,7 @@ def _set_envs_and_config(server_args: ServerArgs):
     if server_args.attention_backend == "flashinfer":
         assert_pkg_version(
             "flashinfer_python",
-            "0.2.1.post1",
+            "0.2.1.post2",
             "Please uninstall the old version and "
             "reinstall the latest version by following the instructions "
             "at https://docs.flashinfer.ai/installation.html.",

sglang/srt/layers/attention/flashinfer_backend.py CHANGED Viewed

@@ -54,7 +54,9 @@ class DecodeMetadata:
 @dataclass
 class PrefillMetadata:
-    prefill_wrappers: List[BatchPrefillWithPagedKVCacheWrapper]
+    prefill_wrappers: List[
+        Union[BatchPrefillWithPagedKVCacheWrapper, BatchMLAPagedAttentionWrapper]
+    ]
     use_ragged: bool
     extend_no_prefix: bool
@@ -160,16 +162,36 @@ class FlashInferAttnBackend(AttentionBackend):
         self.decode_wrappers = []
         for _ in range(self.num_wrappers):
             if not skip_prefill:
-                self.prefill_wrappers_paged.append(
-                    BatchPrefillWithPagedKVCacheWrapper(
-                        self.workspace_buffer,
-                        "NHD",
-                        backend="fa2",
+                if (
+                    self.enable_flashinfer_mla
+                    and not global_server_args_dict["disable_radix_cache"]
+                ):
+                    # use mla paged prefill
+                    self.prefill_wrappers_paged.append(
+                        BatchMLAPagedAttentionWrapper(
+                            self.workspace_buffer,
+                            backend="fa2",
+                        )
+                    )
+                    self.prefill_wrappers_verify.append(
+                        BatchMLAPagedAttentionWrapper(
+                            self.workspace_buffer,
+                            backend="fa2",
+                        )
+                    )
+                else:
+                    self.prefill_wrappers_paged.append(
+                        BatchPrefillWithPagedKVCacheWrapper(
+                            self.workspace_buffer,
+                            "NHD",
+                            backend="fa2",
+                        )
+                    )
+                    self.prefill_wrappers_verify.append(
+                        BatchPrefillWithPagedKVCacheWrapper(
+                            self.workspace_buffer, "NHD"
+                        )
                     )
-                )
-                self.prefill_wrappers_verify.append(
-                    BatchPrefillWithPagedKVCacheWrapper(self.workspace_buffer, "NHD")
-                )
             if self.enable_flashinfer_mla:
                 self.decode_wrappers.append(
                     BatchMLAPagedAttentionWrapper(self.workspace_buffer, backend="fa2")
@@ -237,7 +259,10 @@ class FlashInferAttnBackend(AttentionBackend):
         else:
             prefix_lens = forward_batch.extend_prefix_lens
-            if self.is_multimodal:
+            if self.is_multimodal or (
+                self.enable_flashinfer_mla
+                and not global_server_args_dict["disable_radix_cache"]
+            ):
                 use_ragged = False
                 extend_no_prefix = False
             else:
@@ -419,23 +444,43 @@ class FlashInferAttnBackend(AttentionBackend):
             logits_soft_cap = layer.logit_cap
-            o1, _ = self.prefill_wrapper_ragged.forward_return_lse(
-                q.view(-1, layer.tp_q_head_num, layer.head_dim),
-                k.view(-1, layer.tp_k_head_num, layer.head_dim),
-                v.view(-1, layer.tp_v_head_num, layer.v_head_dim),
-                causal=True,
-                sm_scale=layer.scaling,
-                logits_soft_cap=logits_soft_cap,
-            )
-            o = o1
+            if global_server_args_dict["disable_radix_cache"]:
+                # use mla ragged prefill
+                o, _ = self.prefill_wrapper_ragged.forward_return_lse(
+                    q.view(-1, layer.tp_q_head_num, layer.head_dim),
+                    k.view(-1, layer.tp_k_head_num, layer.head_dim),
+                    v.view(-1, layer.tp_v_head_num, layer.v_head_dim),
+                    causal=True,
+                    sm_scale=layer.scaling,
+                    logits_soft_cap=logits_soft_cap,
+                )
-            if save_kv_cache:
-                forward_batch.token_to_kv_pool.set_kv_buffer(
-                    layer,
-                    cache_loc,
-                    k,
-                    v,
+                if save_kv_cache:
+                    forward_batch.token_to_kv_pool.set_kv_buffer(
+                        layer,
+                        cache_loc,
+                        k,
+                        v,
+                    )
+            else:
+                # use mla paged prefill
+                prefill_wrapper_paged = self.forward_metadata.prefill_wrappers[
+                    self._get_wrapper_idx(layer)
+                ]
+                if k is not None:
+                    assert v is not None
+                    if save_kv_cache:
+                        forward_batch.token_to_kv_pool.set_kv_buffer(
+                            layer, cache_loc, k, v
+                        )
+                qall = q.view(-1, layer.tp_q_head_num, layer.head_dim)
+                k_buf = forward_batch.token_to_kv_pool.get_key_buffer(layer.layer_id)
+                o = prefill_wrapper_paged.run(
+                    qall[:, :, : layer.v_head_dim],
+                    qall[:, :, layer.v_head_dim :],
+                    k_buf[:, :, : layer.v_head_dim],
+                    k_buf[:, :, layer.v_head_dim :],
                 )
             return o.view(-1, layer.tp_q_head_num * layer.v_head_dim)
@@ -800,7 +845,9 @@ class FlashInferIndicesUpdaterPrefill:
         seq_lens: torch.Tensor,
         seq_lens_sum: int,
         prefix_lens: torch.Tensor,
-        prefill_wrappers: List[BatchPrefillWithPagedKVCacheWrapper],
+        prefill_wrappers: List[
+            Union[BatchPrefillWithPagedKVCacheWrapper, BatchMLAPagedAttentionWrapper]
+        ],
         use_ragged: bool,
         encoder_lens: Optional[torch.Tensor],
         spec_info: Optional[SpecInfo],
@@ -814,7 +861,9 @@ class FlashInferIndicesUpdaterPrefill:
         seq_lens: torch.Tensor,
         seq_lens_sum: int,
         prefix_lens: torch.Tensor,
-        prefill_wrappers: List[BatchPrefillWithPagedKVCacheWrapper],
+        prefill_wrappers: List[
+            Union[BatchPrefillWithPagedKVCacheWrapper, BatchMLAPagedAttentionWrapper]
+        ],
         use_ragged: bool,
         encoder_lens: Optional[torch.Tensor],
         spec_info: Optional[SpecInfo],
@@ -923,7 +972,9 @@ class FlashInferIndicesUpdaterPrefill:
     def call_begin_forward(
         self,
         wrapper_ragged: BatchPrefillWithRaggedKVCacheWrapper,
-        wrapper_paged: BatchPrefillWithPagedKVCacheWrapper,
+        wrapper_paged: Union[
+            BatchPrefillWithPagedKVCacheWrapper, BatchMLAPagedAttentionWrapper
+        ],
         req_pool_indices: torch.Tensor,
         paged_kernel_lens: torch.Tensor,
         paged_kernel_lens_sum: int,
@@ -1004,6 +1055,26 @@ class FlashInferIndicesUpdaterPrefill:
                 custom_mask=custom_mask,
                 non_blocking=True,
             )
+        elif (
+            global_config.enable_flashinfer_mla
+            and not global_server_args_dict["disable_radix_cache"]
+        ):
+            # mla paged prefill
+            kv_len_arr = kv_indptr[1:] - kv_indptr[:-1]
+            wrapper_paged.plan(
+                qo_indptr,
+                kv_indptr,
+                kv_indices,
+                kv_len_arr,
+                self.num_qo_heads,
+                512,
+                64,
+                1,
+                True,
+                1 / math.sqrt(192),
+                self.data_type,
+                self.data_type,
+            )
 class FlashInferMultiStepDraftBackend:

sglang/srt/layers/moe/fused_moe_triton/configs/E=256,N=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,146 @@
+{
+    "1": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "2": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "8": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "16": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "24": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "32": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "48": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "64": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "96": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "128": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "256": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "512": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "1024": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "1536": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=1536,K=1536,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 2
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=256,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 5
+    }
+}

sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 2
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2
+    }
+}

sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 2
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 2
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2
+    }
+}

sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json ADDED Viewed

@@ -0,0 +1,26 @@
+{
+    "2048": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 4,
+        "num_stages": 3
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 64,
+        "num_warps": 4,
+        "num_stages": 3
+    }
+}

sglang/srt/managers/schedule_batch.py CHANGED Viewed

@@ -66,6 +66,7 @@ global_server_args_dict = {
     "enable_ep_moe": ServerArgs.enable_ep_moe,
     "device": ServerArgs.device,
     "enable_flashinfer_mla": ServerArgs.enable_flashinfer_mla,
+    "disable_radix_cache": ServerArgs.disable_radix_cache,
 }
 logger = logging.getLogger(__name__)

sglang/srt/model_executor/model_runner.py CHANGED Viewed

@@ -177,6 +177,7 @@ class ModelRunner:
                 "enable_ep_moe": server_args.enable_ep_moe,
                 "device": server_args.device,
                 "enable_flashinfer_mla": server_args.enable_flashinfer_mla,
+                "disable_radix_cache": server_args.disable_radix_cache,
             }
         )

sglang/srt/models/deepseek_v2.py CHANGED Viewed

@@ -511,8 +511,11 @@ class DeepseekV2AttentionMLA(nn.Module):
         forward_batch: ForwardBatch,
     ) -> torch.Tensor:
         if global_server_args_dict["enable_flashinfer_mla"]:
-            if forward_batch.forward_mode.is_extend():
-                return self.forward_normal(positions, hidden_states, forward_batch)
+            if global_server_args_dict["disable_radix_cache"]:
+                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)
         else:

sglang/version.py CHANGED Viewed

	@@ -1 +1 @@
1	- __version__ = "0.4.3.~~post1~~"
1	+ __version__ = "0.4.3.post2"

{sglang-0.4.3.post1.dist-info → sglang-0.4.3.post2.dist-info}/METADATA RENAMED Viewed

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: sglang
-Version: 0.4.3.post1
+Version: 0.4.3.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
@@ -243,7 +243,7 @@ Requires-Dist: cuda-python; 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>=0.2.1.post1; extra == "srt"
+Requires-Dist: flashinfer_python>=0.2.1.post2; 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-0.4.3.post1.dist-info → sglang-0.4.3.post2.dist-info}/RECORD RENAMED Viewed

@@ -10,7 +10,7 @@ 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=9fm5ghtYPXqsWKjUzlQKJIoH5iFit6Rz21RhyaC3YL4,15673
-sglang/version.py,sha256=rH9jaCKrx1Ahm1bUadSFX0yjfqoKnuKVlVyraMi28AU,28
+sglang/version.py,sha256=AJcJAUlaCr4igIr4vVaRiJeU678q5BdTdso-33Tq98k,28
 sglang/lang/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 sglang/lang/chat_template.py,sha256=0tZX67LgtYGrWopnSuTeqWVdxaw2deJOFWOBJpd6htU,17547
 sglang/lang/choices.py,sha256=-W1DVw9N9ZliVpvmWrzIXG4cswAah8eMQrHWzkS3D8o,6234
@@ -61,7 +61,7 @@ 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=cEVosKgOTKF8dKX7wA1vaVOdUP0qjFlZ-X9I4PJ_Ta0,17555
+sglang/srt/entrypoints/engine.py,sha256=t2UhSOnr22BBCyoIhDJZIcmthQOQcL1iaB06LZqqAnU,17555
 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
@@ -77,7 +77,7 @@ 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=8bEnIXFXBB8lv0ruH_paqT8uLIHsKYKYhQ9ytzDcsYI,47681
+sglang/srt/layers/attention/flashinfer_backend.py,sha256=MnQCTvOOPXy4RXvvZNSMjm1nPHXoysQ1cWGVf_yilQU,50518
 sglang/srt/layers/attention/torch_native_backend.py,sha256=KrcAqTLVZLtwgOmB0xhwUUsX32M-5LYZpNxaRNT4VuA,9252
 sglang/srt/layers/attention/triton_backend.py,sha256=mbYaYKHYrUyL2zEXrPtDRIcvVNe6L-bmcdLhKF92V-0,21292
 sglang/srt/layers/attention/vision.py,sha256=zLjKmzUlkgq1RFcP3b4EPArOAKovoaDLgYfM5SyB2wM,13181
@@ -124,6 +124,7 @@ sglang/srt/layers/moe/fused_moe_triton/layer.py,sha256=RWBo3j5AzZls5eD2eaejetSfM
 "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=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=XmKFaMheq7NNrsvYCJteul0w809l_l460ZiDQC9ToGs,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=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
@@ -210,24 +211,30 @@ sglang/srt/layers/quantization/modelopt_quant.py,sha256=_VdVz77dTP-IczPeFrdH6Ttr
 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=1536,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9qdyh6ki9LAyq7VDO9WMRmBOPWKSrZhU-I7z1E9bTKA,550
 "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_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
+"sglang/srt/layers/quantization/configs/N=1536,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=6POXzQZHdNwcBDv1w6BJKbLMRDt0jbFUuMsMNf-ToEs,549
 "sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=7v4tp0RaT4vxF4urSBrkK5FR_5ikeFQ1htF3DwDl1lk,3249
 "sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=3matoCzEe4aexwoe7YTmkjyE4NA8khWXjL5EySuNwzA,3254
 "sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0J2MFgaLkv-mfVE5x363lgVKYU6miLG_xRO3tJUga_M,3249
+"sglang/srt/layers/quantization/configs/N=2048,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=JeXNLkbMAjdDKV-WpzQy87SXN06towo3xUofLtvYCQI,551
 "sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=4ubbhwSFX_XbefRLEkLoWxJkcetFWPzsszPu0X3_Wrw,3242
 "sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9f8Ib4gLEFSfdNpO8IL8uiONImvqnlPbJrZ0HM3OB-o,3247
 "sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FhyniGTx5QeCuVrBSVTQys6q05Pr5lPEcPykpAX7Iyo,3247
+"sglang/srt/layers/quantization/configs/N=2304,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Boa83ZSPZ0LvzqtfLGvois5QK4TmJfwjA2n96c9ET58,549
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0v17v78pETXv6S2ZoibekxOVhiTmCm807DYG4DONUck,3259
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=I44PvJj758-sw_fCOVROLTpG0NQ5_5PCYyQcpZC1YSY,3259
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=tG5_iVeRBHTgHX-liOf79nWRjj_lUZ-NQWTbBrBgORQ,3246
 "sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=-CVHqClROli9FWe_FnlnuAG2LiFivDFK_nghH6t-BWc,3261
+"sglang/srt/layers/quantization/configs/N=24576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=9tCZxJ0eAD7AYMH7OqS3AGppJUllKnJLNvMq7FMXdsA,552
 "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=256,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=QMVfMXS0Yjgob8_9xps1xuZi6KnY5l2MeKxXLRjTeg4,548
 "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_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
@@ -240,6 +247,7 @@ sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id
 "sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Mtw7a9BSspj2TzC-aPxE82o1LEvwzgbUuIofwRxUNA0,3263
 "sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=B0lo3SuoQXhBEnojH2TwpVeurvlKD8yI8kQrJ5ORhWU,3249
 "sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=NHdx3tZnfLF7NplswMzcTRbQEQFLtChg4rd7GU9lMbM,3262
+"sglang/srt/layers/quantization/configs/N=32768,K=512,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ZYZ03ziPGwG2sCasEYLj7ZIP7vNO8UNBR5qNTmKgRMs,549
 "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
@@ -266,22 +274,28 @@ sglang/srt/layers/quantization/w8a8_int8.py,sha256=PBapNo27rkUdX95E0ihUZeHTXP2Id
 "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
 "sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=NiorJgOotxkQcP49ID3z5al1UA4QQDrT8MvbCwAWL5Y,3248
+"sglang/srt/layers/quantization/configs/N=576,K=7168,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FcuzcmKFf2RbaUpAaAsuObUefcGMgNPMDbVdHXRkoGY,549
 "sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=bPQWtvaJrzOOIgI-R-MIxs_f4yC_FobkDydu3OkOFtg,3252
 "sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=h32lCTFilLlyKbMeuJvNWG1v0yJJzNj93kwSvlrHfaY,3249
 "sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ZRgiuHZ2SFC6u-WV5DGwau4k1RiPLI67eENO0e-5Ylg,3253
+"sglang/srt/layers/quantization/configs/N=7168,K=1024,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=kjQ_kvF38bZGcmaeJGSJsSR0NcUjUOh3LZ2-5c4kPvE,550
 "sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=-hP_P8NM0K04mGzTmpGBNibQ5xxh5gPz5WtoMXhoz1E,3253
 "sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=0jX-z2lTgVw7ABLmWsIsQdqW4EjmbXKRDHye_XPLCAE,3245
 "sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=FB5Le4obvPoCgFSnC_3-Uh59n-Mt4Rol8saXVcK3RPw,3252
+"sglang/srt/layers/quantization/configs/N=7168,K=1152,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=mZGU39sobtUqNYKjtyIGjhOZyCOQFJMF3MinA1zjTJA,550
 "sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=kLviGvVngpgOuelfKtvv9Is7MWQ89rGxlomMRP6t0Ic,3250
 "sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=_exM3wJ3FMmGHweBcH-8IxwZBzaOmPaF3ScMM6KDpiY,3253
+"sglang/srt/layers/quantization/configs/N=7168,K=128,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ue2oWml2ouUTZelYx5Nt5pgCmY-ib3mLV1reJL9ZudE,550
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=vlys0Zi_CaaU41OHGbWSBtbVglFi98bgqEySBMc9Sdg,3258
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=YWyByOlKSqp5lbcUa8eu6N2dHRKJqJDbCDSjdDQJngg,3249
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H20,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ENRWYdUwI0ooHb6IwcHliupRWOPnw-7-WtxZB-qQGJI,3245
 "sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_H200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=j5PTW0IC4Z2yQIygcdICaOsvb639u6Mv-ZpJYkrBQ2k,3254
+"sglang/srt/layers/quantization/configs/N=7168,K=16384,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Nv9KP_KLGsRJdJF755dZBvbTws37u1GM2UigMRlAtl0,552
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_B200,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Ggy4hejkcWjiw5Bi-wGzSP5JLVuvOjip_rbjXFBJZbs,3257
 "sglang/srt/layers/quantization/configs/N=7168,K=18432,device_name=NVIDIA_H100_80GB_HBM3,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=Xy4mgZx5iiEvuv2ydO4dFNIT8s0jgBhNHE1vu93fGJM,3250
 "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=18432,device_name=NVIDIA_L20Y,dtype=fp8_w8a8,block_shape=[128, 128].json",sha256=ZOWEo__oUy8AhJiAlRCuGNAZNdNweFdWBFptJYkwxs8,552
 "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_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
@@ -320,7 +334,7 @@ sglang/srt/managers/data_parallel_controller.py,sha256=b64aC6iLr5RolJyNQnT-yTQ_T
 sglang/srt/managers/detokenizer_manager.py,sha256=XC2INyykOgwmIrFEGc-zf6LGZ5mMt6oPZt1YRXW_cbY,9650
 sglang/srt/managers/image_processor.py,sha256=AWtCjl_zCbcn5LD4Hp4NXmsu225lQE0gWixIhQuUMpE,23872
 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_batch.py,sha256=70smg65ed4xbotRegd8NzRu1YKA6M0oKn2Q69i0qWgA,49246
 sglang/srt/managers/schedule_policy.py,sha256=Qero_lwPEb7bM87qjWtYijGyRhtY0mMwjWP6SbjvaUE,18260
 sglang/srt/managers/scheduler.py,sha256=w0FPjiU5MoyP58UdJoPBr-hf-WmlWPpqb-5TSJDJBLo,71908
 sglang/srt/managers/session_controller.py,sha256=WXRbtninVEVM0rQYiXFzOwsDph0TNj1L2sRCWQF0dSg,5571
@@ -337,7 +351,7 @@ sglang/srt/metrics/collector.py,sha256=_yl0_paSARxS1ypZgd-pLJ29tMizolHuwROX21dOX
 sglang/srt/metrics/func_timer.py,sha256=VFyNRrbnKVCwnQsrlLin1lITJfjQpf9m8sGPqL5LIsQ,3438
 sglang/srt/model_executor/cuda_graph_runner.py,sha256=hH646E_c4UlclGEawPDjg4KHgTUEk70WrPl6C7nnltM,18774
 sglang/srt/model_executor/forward_batch_info.py,sha256=cTyRuJVBTBmkP4LAfScRSRrpjLCq7UfmUKoXuU5LZUw,15098
-sglang/srt/model_executor/model_runner.py,sha256=uohQ2n2R1HcVyaHwbdwM6xDvFxZSLgxacjMSrrogLpw,33537
+sglang/srt/model_executor/model_runner.py,sha256=FNi5BTrpQ5VQ-VsMHujaAuSno3Y3DYYjymzPJLJoOIM,33609
 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
@@ -348,7 +362,7 @@ sglang/srt/models/commandr.py,sha256=y8DFUW0NKbkoY2DP6nhgJ1f7F_ysjaHEkEnZYZW2zdk
 sglang/srt/models/dbrx.py,sha256=-L9QkUr_xuMuI6mn0AzG_VE1MqRXoaaFtD4r8UuAzkY,14789
 sglang/srt/models/deepseek.py,sha256=KfcQ54BqlS73XQmtcG0sfnmm3VXOGwUIkd34WS6Gp0Y,15694
 sglang/srt/models/deepseek_nextn.py,sha256=QmzByVDFw8F5cJfBU4-VVryXovn4HxvGBwbBTfJavJg,11740
-sglang/srt/models/deepseek_v2.py,sha256=Er72pYPVxs6hpms9yJL4iSQou7J6kA7mCsmapX9_LJQ,39248
+sglang/srt/models/deepseek_v2.py,sha256=vjY8BznqlEjX4P_mZQp1Syv24YxVee9Q258O8KWqE8I,39429
 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
@@ -421,8 +435,8 @@ 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-0.4.3.post1.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
-sglang-0.4.3.post1.dist-info/METADATA,sha256=TypZMxQ7xbJ3Xh34H0HYZV4bZ8qrID2KMbtggp7j3mQ,23821
-sglang-0.4.3.post1.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
-sglang-0.4.3.post1.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
-sglang-0.4.3.post1.dist-info/RECORD,,
+sglang-0.4.3.post2.dist-info/LICENSE,sha256=FJXh51fvTQklojUFY89XVLsjxRcBqOxPs8XNy-2uZ0c,11346
+sglang-0.4.3.post2.dist-info/METADATA,sha256=kChnxBYuvq-HiSlLlTj6l3bOcIj_mLFuOsSQpUMGkZE,23821
+sglang-0.4.3.post2.dist-info/WHEEL,sha256=In9FTNxeP60KnTkGw7wk6mJPYd_dQSjEZmXdBdMCI-8,91
+sglang-0.4.3.post2.dist-info/top_level.txt,sha256=yxhh3pYQkcnA7v3Bg889C2jZhvtJdEincysO7PEB09M,7
+sglang-0.4.3.post2.dist-info/RECORD,,

{sglang-0.4.3.post1.dist-info → sglang-0.4.3.post2.dist-info}/LICENSE RENAMED Viewed

File without changes

{sglang-0.4.3.post1.dist-info → sglang-0.4.3.post2.dist-info}/WHEEL RENAMED Viewed

File without changes

{sglang-0.4.3.post1.dist-info → sglang-0.4.3.post2.dist-info}/top_level.txt RENAMED Viewed

File without changes

sglang 0.4.3.post1__py3-none-any.whl → 0.4.3.post2__py3-none-any.whl

sglang 0.4.3.post1py3-none-any.whl → 0.4.3.post2py3-none-any.whl