sglang 0.3.3__py3-none-any.whl → 0.3.4__py3-none-any.whl

sglang/srt/layers/attention/flashinfer_backend.py

@@ -7,18 +7,17 @@ FlashInfer is faster and Triton is easier to customize.
 Each backend supports two operators: extend (i.e. prefill with cached prefix) and decode.
 """
 
+from enum import Enum, auto
 from typing import TYPE_CHECKING
 
 import torch
 import torch.nn as nn
+import triton
+import triton.language as tl
 
 from sglang.global_config import global_config
 from sglang.srt.layers.attention import AttentionBackend
-from sglang.srt.layers.attention.flashinfer_utils import (
-    WrapperDispatch,
-    update_flashinfer_indices,
-)
-from sglang.srt.model_executor.forward_batch_info import ForwardBatch, ForwardMode
+from sglang.srt.model_executor.forward_batch_info import ForwardBatch
 from sglang.srt.utils import is_flashinfer_available
 
 if TYPE_CHECKING:
@@ -34,13 +33,18 @@ if is_flashinfer_available():
     from flashinfer.decode import _grouped_size_compiled_for_decode_kernels
 
 
+class WrapperDispatch(Enum):
+    SLIDING_WINDOW = auto()
+    CROSS_ATTENTION = auto()
+
+
 class FlashInferAttnBackend(AttentionBackend):
     """Flashinfer attention kernels."""
 
     def __init__(self, model_runner: ModelRunner):
         super().__init__()
-        self.model_runner = model_runner
 
+        # Parse constants
         if not _grouped_size_compiled_for_decode_kernels(
             model_runner.model_config.num_attention_heads // model_runner.tp_size,
             model_runner.model_config.get_num_kv_heads(model_runner.tp_size),
@@ -48,27 +52,43 @@ class FlashInferAttnBackend(AttentionBackend):
             self.decode_use_tensor_cores = True
         else:
             self.decode_use_tensor_cores = False
-
-        self.workspace_buffer = torch.empty(
-            global_config.flashinfer_workspace_size,
-            dtype=torch.uint8,
-            device="cuda",
-        )
+        self.max_context_len = model_runner.model_config.context_len
 
         assert not (
             model_runner.sliding_window_size is not None
             and model_runner.has_cross_attention
         ), "Sliding window and cross attention are not supported together"
 
-        self.num_wrappers = 1
-        self.dispatch_reason = None
         if model_runner.sliding_window_size is not None:
             self.num_wrappers = 2
             self.dispatch_reason = WrapperDispatch.SLIDING_WINDOW
         elif model_runner.has_cross_attention:
             self.num_wrappers = 2
             self.dispatch_reason = WrapperDispatch.CROSS_ATTENTION
+        else:
+            self.num_wrappers = 1
+            self.dispatch_reason = None
+
+        # Allocate buffers
+        self.workspace_buffer = torch.empty(
+            global_config.flashinfer_workspace_size,
+            dtype=torch.uint8,
+            device=model_runner.device,
+        )
+        max_bs = model_runner.req_to_token_pool.size
+        self.kv_indptr = [
+            torch.zeros((max_bs + 1,), dtype=torch.int32, device=model_runner.device)
+            for _ in range(self.num_wrappers)
+        ]
+        self.kv_last_page_len = torch.ones(
+            (max_bs,), dtype=torch.int32, device=model_runner.device
+        )
+        self.qo_indptr = [
+            torch.zeros((max_bs + 1,), dtype=torch.int32, device=model_runner.device)
+            for _ in range(self.num_wrappers)
+        ]
 
+        # Create wrappers
         # NOTE: we do not use ragged attention when there are multiple wrappers
         self.prefill_wrapper_ragged = (
             BatchPrefillWithRaggedKVCacheWrapper(self.workspace_buffer, "NHD")
@@ -92,26 +112,23 @@ class FlashInferAttnBackend(AttentionBackend):
                 )
             )
 
+        # Create indices updater
+        self.indices_updater_decode = FlashInferIndicesUpdaterDecode(model_runner, self)
+        self.indices_updater_prefill = FlashInferIndicesUpdaterPrefill(
+            model_runner, self
+        )
+
+        # Other metadata
         self.forward_metadata = None
         self.cuda_graph_metadata = {}
 
-    def _get_wrapper_idx(self, layer: nn.Module):
-        if self.num_wrappers == 1:
-            return 0
-
-        if self.dispatch_reason == WrapperDispatch.SLIDING_WINDOW:
-            return layer.sliding_window_size == -1
-        if self.dispatch_reason == WrapperDispatch.CROSS_ATTENTION:
-            return layer.is_cross_attention
-
-        raise ValueError(f"Unknown dispatch reason: {self.dispatch_reason}")
-
     def init_forward_metadata(self, forward_batch: ForwardBatch):
         if forward_batch.forward_mode.is_decode():
-            prefix_lens = None
-            use_ragged = False
-            extend_no_prefix = False
-            total_num_tokens = None
+            self.indices_updater_decode.update(
+                forward_batch.req_pool_indices,
+                forward_batch.seq_lens,
+            )
+            self.forward_metadata = (self.decode_wrappers,)
         else:
             prefix_lens = forward_batch.extend_prefix_lens
 
@@ -123,48 +140,32 @@ class FlashInferAttnBackend(AttentionBackend):
             ):
                 use_ragged = True
 
-            total_num_tokens = torch.sum(forward_batch.seq_lens).item()
             extend_no_prefix = not torch.any(forward_batch.extend_prefix_lens).item()
 
-        update_flashinfer_indices(
-            forward_batch.forward_mode,
-            self.model_runner,
-            forward_batch.req_pool_indices,
-            forward_batch.seq_lens,
-            prefix_lens,
-            use_ragged=use_ragged,
-        )
+            self.indices_updater_prefill.update(
+                forward_batch.req_pool_indices,
+                forward_batch.seq_lens,
+                prefix_lens,
+                use_ragged,
+            )
 
-        self.forward_metadata = (
-            use_ragged,
-            extend_no_prefix,
-            total_num_tokens,
-            self.decode_wrappers,
-        )
+            self.forward_metadata = (
+                use_ragged,
+                extend_no_prefix,
+            )
 
     def init_cuda_graph_state(self, max_bs: int):
-        self.cuda_graph_kv_indptr = torch.zeros(
-            (max_bs + 1,), dtype=torch.int32, device="cuda"
-        )
-        self.cuda_graph_kv_indices = torch.zeros(
-            (max_bs * self.model_runner.model_config.context_len,),
+        cuda_graph_kv_indices = torch.zeros(
+            (max_bs * self.max_context_len,),
             dtype=torch.int32,
             device="cuda",
         )
-        self.cuda_graph_kv_last_page_len = torch.ones(
-            (max_bs,), dtype=torch.int32, device="cuda"
-        )
-
-        # NOTE: the buffers are always in the form of list
-        self.cuda_graph_kv_indptr = [self.cuda_graph_kv_indptr] + [
-            self.cuda_graph_kv_indptr.clone() for _ in range(self.num_wrappers - 1)
-        ]
-        self.cuda_graph_kv_indices = [self.cuda_graph_kv_indices] + [
-            self.cuda_graph_kv_indices.clone() for _ in range(self.num_wrappers - 1)
+        self.cuda_graph_kv_indices = [cuda_graph_kv_indices] + [
+            cuda_graph_kv_indices.clone() for _ in range(self.num_wrappers - 1)
         ]
 
     def init_forward_metadata_capture_cuda_graph(
-        self, bs: int, req_pool_indices, seq_lens
+        self, bs: int, req_pool_indices: torch.Tensor, seq_lens: torch.Tensor
     ):
         decode_wrappers = []
         for i in range(self.num_wrappers):
@@ -174,35 +175,21 @@ class FlashInferAttnBackend(AttentionBackend):
                     "NHD",
                     use_cuda_graph=True,
                     use_tensor_cores=self.decode_use_tensor_cores,
-                    paged_kv_indptr_buffer=self.cuda_graph_kv_indptr[i][: bs + 1],
+                    paged_kv_indptr_buffer=self.kv_indptr[i][: bs + 1],
                     paged_kv_indices_buffer=self.cuda_graph_kv_indices[i],
-                    paged_kv_last_page_len_buffer=self.cuda_graph_kv_last_page_len[:bs],
+                    paged_kv_last_page_len_buffer=self.kv_last_page_len[:bs],
                 )
             )
 
-        update_flashinfer_indices(
-            ForwardMode.DECODE,
-            self.model_runner,
-            req_pool_indices,
-            seq_lens,
-            None,
-            decode_wrappers,
-        )
-
+        self.indices_updater_decode.update(req_pool_indices, seq_lens, decode_wrappers)
         self.cuda_graph_metadata[bs] = decode_wrappers
-
-        self.forward_metadata = (False, False, None, decode_wrappers)
+        self.forward_metadata = (decode_wrappers,)
 
     def init_forward_metadata_replay_cuda_graph(
-        self, bs: int, req_pool_indices, seq_lens
+        self, bs: int, req_pool_indices: torch.Tensor, seq_lens: torch.Tensor
     ):
-        update_flashinfer_indices(
-            ForwardMode.DECODE,
-            self.model_runner,
-            req_pool_indices[:bs],
-            seq_lens[:bs],
-            None,
-            self.cuda_graph_metadata[bs],
+        self.indices_updater_decode.update(
+            req_pool_indices[:bs], seq_lens[:bs], self.cuda_graph_metadata[bs]
         )
 
     def get_cuda_graph_seq_len_fill_value(self):
@@ -213,7 +200,7 @@ class FlashInferAttnBackend(AttentionBackend):
             self._get_wrapper_idx(layer)
         ]
 
-        use_ragged, extend_no_prefix, _, _ = self.forward_metadata
+        use_ragged, extend_no_prefix = self.forward_metadata
 
         if not use_ragged:
             if k is not None:
@@ -259,7 +246,7 @@ class FlashInferAttnBackend(AttentionBackend):
         return o.view(-1, layer.tp_q_head_num * layer.head_dim)
 
     def forward_decode(self, q, k, v, layer: nn.Module, forward_batch: ForwardBatch):
-        decode_wrapper = self.forward_metadata[-1][self._get_wrapper_idx(layer)]
+        decode_wrapper = self.forward_metadata[0][self._get_wrapper_idx(layer)]
 
         if k is not None:
             assert v is not None
@@ -275,3 +262,285 @@ class FlashInferAttnBackend(AttentionBackend):
         )
 
         return o.view(-1, layer.tp_q_head_num * layer.head_dim)
+
+    def _get_wrapper_idx(self, layer: nn.Module):
+        if self.num_wrappers == 1:
+            return 0
+
+        if self.dispatch_reason == WrapperDispatch.SLIDING_WINDOW:
+            return layer.sliding_window_size == -1
+        if self.dispatch_reason == WrapperDispatch.CROSS_ATTENTION:
+            return layer.is_cross_attention
+
+        raise ValueError(f"Unknown dispatch reason: {self.dispatch_reason}")
+
+
+class FlashInferIndicesUpdaterDecode:
+    def __init__(self, model_runner: ModelRunner, attn_backend: AttentionBackend):
+        # Constants
+        self.num_qo_heads = (
+            model_runner.model_config.num_attention_heads // model_runner.tp_size
+        )
+        self.num_kv_heads = model_runner.model_config.get_num_kv_heads(
+            model_runner.tp_size
+        )
+        self.head_dim = model_runner.model_config.head_dim
+        self.data_type = model_runner.kv_cache_dtype
+        self.q_data_type = model_runner.dtype
+        self.max_context_len = model_runner.req_to_token_pool.req_to_token.size(1)
+        self.sliding_window_size = model_runner.sliding_window_size
+
+        # Buffers and wrappers
+        self.kv_indptr = attn_backend.kv_indptr
+        self.kv_last_page_len = attn_backend.kv_last_page_len
+        self.req_to_token = model_runner.req_to_token_pool.req_to_token
+        self.decode_wrappers = attn_backend.decode_wrappers
+
+        # Dispatch
+        if attn_backend.dispatch_reason == WrapperDispatch.SLIDING_WINDOW:
+            self.update = self.update_sliding_window
+        elif attn_backend.dispatch_reason == WrapperDispatch.CROSS_ATTENTION:
+            self.update = self.update_cross_attention
+        else:
+            assert attn_backend.num_wrappers == 1
+            self.update = self.update_single_wrapper
+
+    def update_single_wrapper(self, req_pool_indices, seq_lens, decode_wrappers=None):
+        decode_wrappers = decode_wrappers or self.decode_wrappers
+        self.call_begin_forward(
+            decode_wrappers[0], req_pool_indices, seq_lens, self.kv_indptr[0], None
+        )
+
+    def update_sliding_window(self, req_pool_indices, seq_lens, decode_wrappers=None):
+        decode_wrappers = decode_wrappers or self.decode_wrappers
+
+        for wrapper_id in range(2):
+            if wrapper_id == 0:
+                # Sliding window attention
+                paged_kernel_lens = torch.minimum(  # TODO: replace this with clamp
+                    seq_lens,
+                    torch.tensor(self.sliding_window_size + 1),
+                )
+            else:
+                # Full attention
+                paged_kernel_lens = seq_lens
+
+            kv_start_idx = seq_lens - paged_kernel_lens
+
+            self.call_begin_forward(
+                decode_wrappers[wrapper_id],
+                req_pool_indices,
+                paged_kernel_lens,
+                self.kv_indptr[wrapper_id],
+                kv_start_idx,
+            )
+
+    def update_cross_attention(self):
+        raise NotImplementedError()
+
+    def call_begin_forward(
+        self, wrapper, req_pool_indices, paged_kernel_lens, kv_indptr, kv_start_idx
+    ):
+        bs = len(req_pool_indices)
+        kv_indptr = kv_indptr[: bs + 1]
+        # TODO: optimize the blocking call on kv_indptr[-1]
+        kv_indptr[1:] = torch.cumsum(paged_kernel_lens, dim=0)
+        kv_indices = torch.empty(kv_indptr[-1], dtype=torch.int32, device="cuda")
+
+        create_flashinfer_kv_indices_triton[(bs,)](
+            self.req_to_token,
+            req_pool_indices,
+            paged_kernel_lens,
+            kv_indptr,
+            kv_start_idx,
+            kv_indices,
+            self.max_context_len,
+        )
+
+        wrapper.end_forward()
+        wrapper.begin_forward(
+            kv_indptr,
+            kv_indices,
+            self.kv_last_page_len[:bs],
+            self.num_qo_heads,
+            self.num_kv_heads,
+            self.head_dim,
+            1,
+            data_type=self.data_type,
+            q_data_type=self.q_data_type,
+        )
+
+
+class FlashInferIndicesUpdaterPrefill:
+    def __init__(self, model_runner: ModelRunner, attn_backend: AttentionBackend):
+        # Constants
+        self.num_qo_heads = (
+            model_runner.model_config.num_attention_heads // model_runner.tp_size
+        )
+        self.num_kv_heads = model_runner.model_config.get_num_kv_heads(
+            model_runner.tp_size
+        )
+        self.head_dim = model_runner.model_config.head_dim
+        self.data_type = model_runner.kv_cache_dtype
+        self.q_data_type = model_runner.dtype
+        self.max_context_len = model_runner.req_to_token_pool.req_to_token.size(1)
+        self.sliding_window_size = model_runner.sliding_window_size
+
+        # Buffers and wrappers
+        self.kv_indptr = attn_backend.kv_indptr
+        self.kv_last_page_len = attn_backend.kv_last_page_len
+        self.qo_indptr = attn_backend.qo_indptr
+        self.req_to_token = model_runner.req_to_token_pool.req_to_token
+        self.wrapper_ragged = attn_backend.prefill_wrapper_ragged
+        self.wrappers_paged = attn_backend.prefill_wrappers_paged
+
+        # Dispatch
+        if attn_backend.dispatch_reason == WrapperDispatch.SLIDING_WINDOW:
+            self.update = self.update_sliding_window
+        elif attn_backend.dispatch_reason == WrapperDispatch.CROSS_ATTENTION:
+            self.update = self.update_cross_attention
+        else:
+            assert attn_backend.num_wrappers == 1
+            self.update = self.update_single_wrapper
+
+    def update_single_wrapper(
+        self, req_pool_indices, seq_lens, prefix_lens, use_ragged
+    ):
+        if use_ragged:
+            paged_kernel_lens = prefix_lens
+        else:
+            paged_kernel_lens = seq_lens
+
+        self.call_begin_forward(
+            self.wrapper_ragged,
+            self.wrappers_paged[0],
+            req_pool_indices,
+            paged_kernel_lens,
+            seq_lens,
+            prefix_lens,
+            None,
+            self.kv_indptr[0],
+            self.qo_indptr[0],
+            use_ragged,
+        )
+
+    def update_sliding_window(
+        self, req_pool_indices, seq_lens, prefix_lens, use_ragged
+    ):
+        for wrapper_id in range(2):
+            if wrapper_id == 0:
+                # window attention use paged only
+                paged_kernel_lens = torch.minimum(
+                    seq_lens,
+                    torch.tensor(self.sliding_window_size) + seq_lens - prefix_lens,
+                )
+            else:
+                # full attention
+                paged_kernel_lens = seq_lens
+            kv_start_idx = seq_lens - paged_kernel_lens
+
+            self.call_begin_forward(
+                self.wrapper_ragged,
+                self.wrappers_paged[wrapper_id],
+                req_pool_indices,
+                paged_kernel_lens,
+                seq_lens,
+                prefix_lens,
+                kv_start_idx,
+                self.kv_indptr[wrapper_id],
+                self.qo_indptr[wrapper_id],
+                use_ragged,
+            )
+
+    def update_cross_attention(self):
+        raise NotImplementedError()
+
+    def call_begin_forward(
+        self,
+        wrapper_ragged,
+        wrapper_paged,
+        req_pool_indices,
+        paged_kernel_lens,
+        seq_lens,
+        prefix_lens,
+        kv_start_idx,
+        kv_indptr,
+        qo_indptr,
+        use_ragged,
+    ):
+        bs = len(req_pool_indices)
+        kv_indptr = kv_indptr[: bs + 1]
+        kv_indptr[1:] = torch.cumsum(paged_kernel_lens, dim=0)
+        kv_indices = torch.empty(kv_indptr[-1], dtype=torch.int32, device="cuda")
+        create_flashinfer_kv_indices_triton[(bs,)](
+            self.req_to_token,
+            req_pool_indices,
+            paged_kernel_lens,
+            kv_indptr,
+            kv_start_idx,
+            kv_indices,
+            self.max_context_len,
+        )
+
+        qo_indptr = qo_indptr[: bs + 1]
+        qo_indptr[1:] = torch.cumsum(seq_lens - prefix_lens, dim=0)
+
+        # extend part
+        if use_ragged:
+            wrapper_ragged.end_forward()
+            wrapper_ragged.begin_forward(
+                qo_indptr,
+                qo_indptr,
+                self.num_qo_heads,
+                self.num_kv_heads,
+                self.head_dim,
+            )
+
+        # cached part
+        wrapper_paged.end_forward()
+        wrapper_paged.begin_forward(
+            qo_indptr,
+            kv_indptr,
+            kv_indices,
+            self.kv_last_page_len[:bs],
+            self.num_qo_heads,
+            self.num_kv_heads,
+            self.head_dim,
+            1,
+        )
+
+
+@triton.jit
+def create_flashinfer_kv_indices_triton(
+    req_to_token_ptr,  # [max_batch, max_context_len]
+    req_pool_indices_ptr,
+    page_kernel_lens_ptr,
+    kv_indptr,
+    kv_start_idx,
+    kv_indices_ptr,
+    max_context_len: tl.constexpr,
+):
+    BLOCK_SIZE: tl.constexpr = 512
+    pid = tl.program_id(axis=0)
+    req_pool_index = tl.load(req_pool_indices_ptr + pid)
+    kv_indices_offset = tl.load(kv_indptr + pid)
+
+    kv_start = 0
+    kv_end = 0
+    if kv_start_idx:
+        kv_start = tl.load(kv_start_idx + pid).to(tl.int32)
+        kv_end = kv_start
+    kv_end += tl.load(page_kernel_lens_ptr + pid).to(tl.int32)
+
+    req_to_token_ptr += req_pool_index * max_context_len
+    kv_indices_ptr += kv_indices_offset
+
+    ld_offset = kv_start + tl.arange(0, BLOCK_SIZE)
+    st_offset = tl.arange(0, BLOCK_SIZE)
+    num_loop = tl.cdiv(kv_end - kv_start, BLOCK_SIZE)
+    for _ in range(num_loop):
+        mask = ld_offset < kv_end
+        data = tl.load(req_to_token_ptr + ld_offset, mask=mask)
+        tl.store(kv_indices_ptr + st_offset, data, mask=mask)
+        ld_offset += BLOCK_SIZE
+        st_offset += BLOCK_SIZE

sglang/srt/layers/attention/triton_backend.py

@@ -40,6 +40,8 @@ class TritonAttnBackend(AttentionBackend):
 
         self.cuda_graph_max_seq_len = model_runner.model_config.context_len
 
+        self.device = model_runner.device
+
     def init_forward_metadata(self, forward_batch: ForwardBatch):
         """Init auxiliary variables for triton attention backend."""
 
@@ -51,7 +53,7 @@ class TritonAttnBackend(AttentionBackend):
             attn_logits = torch.empty(
                 (self.num_head, total_num_tokens),
                 dtype=self.reduce_dtype,
-                device="cuda",
+                device=self.device,
             )
 
             max_seq_len = torch.max(forward_batch.seq_lens).item()
@@ -67,7 +69,7 @@ class TritonAttnBackend(AttentionBackend):
         self.cuda_graph_max_total_num_tokens = max_bs * self.cuda_graph_max_seq_len
 
         self.cuda_graph_start_loc = torch.zeros(
-            (max_bs,), dtype=torch.int32, device="cuda"
+            (max_bs,), dtype=torch.int32, device=self.device
         )
         self.cuda_graph_attn_logits = torch.empty(
             (
@@ -79,7 +81,7 @@ class TritonAttnBackend(AttentionBackend):
         )
 
     def init_forward_metadata_capture_cuda_graph(
-        self, bs: int, req_pool_indices, seq_lens
+        self, bs: int, req_pool_indices: torch.Tensor, seq_lens: torch.Tensor
     ):
         self.forward_metadata = (
             self.cuda_graph_start_loc,
@@ -89,7 +91,7 @@ class TritonAttnBackend(AttentionBackend):
         )
 
     def init_forward_metadata_replay_cuda_graph(
-        self, bs: int, req_pool_indices, seq_lens
+        self, bs: int, req_pool_indices: torch.Tensor, seq_lens: torch.Tensor
     ):
         self.cuda_graph_start_loc.zero_()
         self.cuda_graph_start_loc[1:bs] = torch.cumsum(seq_lens[: bs - 1], dim=0)