Merge branch 'v0.9.2-dev-wm' into 'v0.9.2-dev'

[feat]支持mtp模型full_cuda_graph

See merge request dcutoolkit/deeplearing/vllm!172

vllm/model_executor/models/deepseek_mtp.py

@@ -150,7 +150,7 @@ class DeepSeekMultiTokenPredictor(nn.Module):
                                        sampling_metadata)
         return logits
 
-#@support_torch_compile
+@support_torch_compile
 class DeepSeekMTP(nn.Module, SupportsPP):
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):

vllm/v1/attention/backends/mla/common.py

@@ -647,13 +647,13 @@ class MLACommonMetadataBuilder(AttentionMetadataBuilder[M]):
                 rarange = np.repeat(query_lens, query_lens) - arange - 1
 
                 repeats = torch.from_numpy(query_lens).pin_memory().to(
-                    block_table_tensor.device, non_blocking=True)
+                    block_table_tensor.device, non_blocking=True).contiguous()
                 decode_block_table_tensor = torch.repeat_interleave(
                     block_table_tensor[:self._num_decodes, ...],
-                    repeats, dim=0)
-                decode_seq_lens = torch.repeat_interleave(seq_lens[:self._num_decodes], repeats, dim=0)
+                    repeats, dim=0).contiguous()
+                decode_seq_lens = torch.repeat_interleave(seq_lens[:self._num_decodes], repeats, dim=0).contiguous()
                 seq_lens_minus = torch.from_numpy(rarange).to(torch.int32).pin_memory().to(
-                    seq_lens.device, non_blocking=True)
+                    seq_lens.device, non_blocking=True).contiguous()
                 decode_seq_lens = decode_seq_lens - seq_lens_minus
 
                 if self.spec_decode_block_table_tensor is not None:

vllm/v1/spec_decode/eagle.py

 # SPDX-License-Identifier: Apache-2.0
 # SPDX-FileCopyrightText: Copyright contributors to the vLLM project
+from typing import Any, Optional
+import numpy as np
+
 import torch
 import torch.nn as nn
 
@@ -29,10 +32,10 @@ PADDING_SLOT_ID = -1
 class EagleProposer:
 
     def __init__(
-            self,
-            vllm_config: VllmConfig,
-            device: torch.device,
-            runner=None,
+        self,
+        vllm_config: VllmConfig,
+        device: torch.device,
+        runner=None,
     ):
         self.vllm_config = vllm_config
         self.speculative_config = vllm_config.speculative_config
@@ -56,6 +59,9 @@ class EagleProposer:
         self.use_cuda_graph = (self.vllm_config.compilation_config.level
                                == CompilationLevel.PIECEWISE and
                                not self.vllm_config.model_config.enforce_eager)
+        self.use_full_cuda_graph = (
+            self.use_cuda_graph
+            and vllm_config.compilation_config.full_cuda_graph)
         self.cudagraph_batch_sizes = list(
             reversed(
                 self.vllm_config.compilation_config.cudagraph_capture_sizes))
@@ -71,6 +77,9 @@ class EagleProposer:
             (self.max_num_tokens, self.hidden_size),
             dtype=self.dtype,
             device=device)
+
+        # attention metadata captured in full cudagraph mode
+        self.attn_metadata_cudagraph = None
         # We need +1 here because the arange is used to set query_start_loc,
         # which has one more element than batch_size.
         self.arange = torch.arange(vllm_config.scheduler_config.max_num_seqs +
@@ -79,25 +88,25 @@ class EagleProposer:
                                    dtype=torch.int32)
 
     def propose(
-            self,
-            # [num_tokens]
-            target_token_ids: torch.Tensor,
-            # [num_tokens]
-            target_positions: torch.Tensor,
-            # [num_tokens, hidden_size]
-            target_hidden_states: torch.Tensor,
-            # [num_tokens]
-            target_slot_mapping: torch.Tensor,
-            # [batch_size]
-            next_token_ids: torch.Tensor,
-            # [batch_size + 1] starting with 0
-            cu_num_tokens: torch.Tensor,
-            # [batch_size, max_num_blocks_per_req]
-            block_table: torch.Tensor,
-            # [batch_size]
-            num_rejected_tokens: list[int],
-            # [batch_size]
-            sampling_metadata: SamplingMetadata
+        self,
+        # [num_tokens]
+        target_token_ids: torch.Tensor,
+        # [num_tokens]
+        target_positions: torch.Tensor,
+        # [num_tokens, hidden_size]
+        target_hidden_states: torch.Tensor,
+        # [num_tokens]
+        target_slot_mapping: torch.Tensor,
+        # [batch_size]
+        next_token_ids: torch.Tensor,
+        # [batch_size + 1] starting with 0
+        cu_num_tokens: torch.Tensor,
+        # [batch_size, max_num_blocks_per_req]
+        block_table: torch.Tensor,
+        # [batch_size]
+        num_rejected_tokens: list[int],
+        # [batch_size]
+        sampling_metadata: SamplingMetadata
     ) -> torch.Tensor:
         num_tokens = target_token_ids.shape[0]
         batch_size = next_token_ids.shape[0]
@@ -157,7 +166,7 @@ class EagleProposer:
             # FIXME: need to consider multiple kv_cache_groups
             attn_metadata = self.runner.attn_metadata_builders[0].build(
                 common_prefix_len=0,
-                common_attn_metadata=common_attn_metadata,
+                common_attn_metadata=common_attn_metadata
             )
         else:
             raise ValueError(f"Unsupported method: {self.method}")
@@ -168,7 +177,7 @@ class EagleProposer:
         for layer_name in self.attn_layer_names:
             per_layer_attn_metadata[layer_name] = attn_metadata
         if self.use_cuda_graph and \
-                num_tokens <= self.cudagraph_batch_sizes[-1]:
+            num_tokens <= self.cudagraph_batch_sizes[-1]:
             num_input_tokens = self.vllm_config.pad_for_cudagraph(num_tokens)
         else:
             num_input_tokens = num_tokens
@@ -176,6 +185,38 @@ class EagleProposer:
         self.positions[:num_tokens] = target_positions
         self.hidden_states[:num_tokens] = target_hidden_states
 
+        if (self.use_full_cuda_graph
+                and num_tokens <= self.cudagraph_batch_sizes[-1]):
+            assert self.attn_metadata_cudagraph
+            if self.method in ["eagle", "eagle3"]:
+                self.attn_metadata_cudagraph.seq_lens[:batch_size] = (
+                    attn_metadata.seq_lens)
+                self.attn_metadata_cudagraph.slot_mapping[:num_tokens] = (
+                    attn_metadata.slot_mapping)
+                self.attn_metadata_cudagraph.query_start_loc[:batch_size + 1] = (
+                    attn_metadata.query_start_loc)
+                self.attn_metadata_cudagraph.block_table[:batch_size] = (
+                    attn_metadata.block_table)
+            elif self.method == "deepseek_mtp":
+                self.attn_metadata_cudagraph.num_actual_tokens = (
+                    attn_metadata.num_actual_tokens)
+                self.attn_metadata_cudagraph.query_start_loc[:batch_size + 1] = (
+                    attn_metadata.query_start_loc)
+                self.attn_metadata_cudagraph.slot_mapping[:num_tokens] = (
+                    attn_metadata.slot_mapping)
+                self.attn_metadata_cudagraph.num_decodes = (
+                    attn_metadata.num_decodes)
+                self.attn_metadata_cudagraph.num_decode_tokens = (
+                    attn_metadata.num_decode_tokens)
+                self.attn_metadata_cudagraph.num_prefills = (
+                    attn_metadata.num_prefills)
+
+                if attn_metadata.decode is not None:
+                    self.attn_metadata_cudagraph.decode.block_table[:attn_metadata.num_decode_tokens] = (
+                        attn_metadata.decode.block_table)
+                    self.attn_metadata_cudagraph.decode.seq_lens[:attn_metadata.num_decode_tokens] = (
+                        attn_metadata.decode.seq_lens)
+
         with set_forward_context(per_layer_attn_metadata,
                                  self.vllm_config,
                                  num_tokens=num_input_tokens):
@@ -190,6 +231,7 @@ class EagleProposer:
                 last_hidden_states, hidden_states = ret_hidden_states
         sample_hidden_states = last_hidden_states[last_token_indices]
         logits = self.model.compute_logits(sample_hidden_states, None)
+
         draft_token_ids = logits.argmax(dim=-1)
 
         # Early exit if there is only one draft token to be generated.
@@ -227,10 +269,10 @@ class EagleProposer:
             block_table = self.runner.attn_metadata_builders[0].block_table.get_device_tensor()[:batch_size, ...]
             attn_metadata.decode = self.runner.attn_metadata_builders[0]._build_decode(
                 block_table_tensor=block_table,
-                seq_lens=(seq_lens + 1),
+                seq_lens=seq_lens,
             )
 
-        for _ in range(self.num_speculative_tokens - 1):
+        for i in range(self.num_speculative_tokens - 1):
             # Update the inputs.
             # cast to int32 is crucial when eagle model is compiled.
             # tensor.argmax() returns int64 by default.
@@ -267,10 +309,10 @@ class EagleProposer:
             # Compute the slot mapping.
             block_numbers = clamped_positions // self.block_size
             block_ids = block_table.gather(dim=1,
-                                           index=block_numbers.view(-1, 1))
+                                        index=block_numbers.view(-1, 1))
             block_ids = block_ids.view(-1)
             attn_metadata.slot_mapping = (block_ids * self.block_size +
-                                          clamped_positions % self.block_size)
+                                        clamped_positions % self.block_size)
             # Mask out the slot mappings that exceed the max model length.
             # Otherwise, the KV cache will be inadvertently updated with the
             # padding tokens.
@@ -282,6 +324,43 @@ class EagleProposer:
             self.positions[:batch_size] = clamped_positions
             self.hidden_states[:batch_size] = hidden_states
 
+            if (self.use_full_cuda_graph
+                    and batch_size <= self.cudagraph_batch_sizes[-1]):
+                assert self.attn_metadata_cudagraph
+                if self.method in ["eagle", "eagle3"]:
+                    self.attn_metadata_cudagraph.seq_lens[:batch_size] = (
+                        attn_metadata.seq_lens)
+                    self.attn_metadata_cudagraph.slot_mapping[:batch_size] = (
+                        attn_metadata.slot_mapping)
+                    if i == 0:
+                        self.attn_metadata_cudagraph.query_start_loc[:batch_size +
+                                                                    1] = (
+                                                                        attn_metadata
+                                                                        .
+                                                                        query_start_loc
+                                                                    )
+                        self.attn_metadata_cudagraph.block_table[:batch_size] = (
+                            attn_metadata.block_table)
+                elif self.method == "deepseek_mtp":
+                    self.attn_metadata_cudagraph.num_actual_tokens = (
+                        attn_metadata.num_actual_tokens)
+                    self.attn_metadata_cudagraph.slot_mapping[:attn_metadata.num_decode_tokens] = (
+                        attn_metadata.slot_mapping)
+                    self.attn_metadata_cudagraph.num_decodes = (
+                        attn_metadata.num_decodes)
+                    self.attn_metadata_cudagraph.num_decode_tokens = (
+                        attn_metadata.num_decode_tokens)
+                    self.attn_metadata_cudagraph.num_prefills = (
+                        attn_metadata.num_prefills)
+                    self.attn_metadata_cudagraph.decode.seq_lens[:attn_metadata.num_decode_tokens] = (
+                        attn_metadata.decode.seq_lens)
+
+                    if i == 0:
+                        self.attn_metadata_cudagraph.query_start_loc[:batch_size + 1] = (
+                            attn_metadata.query_start_loc)
+                        self.attn_metadata_cudagraph.decode.block_table[:attn_metadata.num_decode_tokens] = (
+                            attn_metadata.decode.block_table)
+
             # Run the model.
             with set_forward_context(per_layer_attn_metadata,
                                      self.vllm_config,
@@ -307,15 +386,16 @@ class EagleProposer:
 
         # [batch_size, num_speculative_tokens]
         draft_token_ids = torch.stack(draft_token_ids_list, dim=1)
+
         return draft_token_ids
 
     @staticmethod
     def prepare_inputs(
-            # [batch_size + 1]
-            cu_target_query_lens: torch.Tensor,
-            # [batch_size]
-            num_rejected_tokens: torch.Tensor,
-            num_tokens: int,
+        # [batch_size + 1]
+        cu_target_query_lens: torch.Tensor,
+        # [batch_size]
+        num_rejected_tokens: torch.Tensor,
+        num_tokens: int,
     ) -> tuple[torch.Tensor, torch.Tensor]:
         # cu_target_query_lens: [0, a, a + b, a + b + c]
         # num_rejected_tokens: [n1, n2, n3]
@@ -342,7 +422,7 @@ class EagleProposer:
         )
         batch_size = num_rejected_tokens.shape[0]
         BLOCK_SIZE = 1024
-        prepare_eagle_input_kernel[(batch_size,)](
+        prepare_eagle_input_kernel[(batch_size, )](
             token_indices,
             cu_target_query_lens,
             cu_num_tokens,
@@ -362,8 +442,8 @@ class EagleProposer:
                                    model_config=draft_model_config)
 
         draft_attn_layer_names = (
-                get_layers_from_vllm_config(self.vllm_config, Attention).keys() -
-                target_attn_layer_names)
+            get_layers_from_vllm_config(self.vllm_config, Attention).keys() -
+            target_attn_layer_names)
 
         self.attn_layer_names = list(draft_attn_layer_names)
 
@@ -376,8 +456,8 @@ class EagleProposer:
             target_language_model = target_model
         # share embed_tokens with the target model if needed
         if get_pp_group().world_size == 1 \
-                and self.method != "deepseek_mtp" \
-                and self.model.model.embed_tokens.weight.shape \
+            and self.method != "deepseek_mtp" \
+            and self.model.model.embed_tokens.weight.shape \
                 == target_language_model.model.embed_tokens.weight.shape:
             logger.info(
                 "Assuming the EAGLE head shares the same vocab embedding" \
@@ -402,10 +482,15 @@ class EagleProposer:
 
     @torch.inference_mode()
     def dummy_run(
-            self,
-            num_tokens: int,
+        self,
+        num_tokens: int,
+        attn_metadata: Optional[dict[str, Any]] = None,
     ) -> None:
-        with set_forward_context(None, self.vllm_config,
+        if attn_metadata is not None and self.attn_metadata_cudagraph is None:
+            self.attn_metadata_cudagraph = attn_metadata[
+                self.attn_layer_names[0]]
+        with set_forward_context(attn_metadata,
+                                 self.vllm_config,
                                  num_tokens=num_tokens):
             self.model(
                 self.input_ids[:num_tokens],
@@ -440,8 +525,8 @@ class EagleProposer:
 # FIXME(woosuk): The logic here is duplicated with the main sampling code.
 # We should refactor this to reuse the same sampling implementation.
 def compute_probs_and_sample_next_token(
-        logits: torch.Tensor,
-        sampling_metadata: SamplingMetadata,
+    logits: torch.Tensor,
+    sampling_metadata: SamplingMetadata,
 ) -> tuple[torch.Tensor, torch.Tensor]:
     if sampling_metadata.all_greedy:
         # For greedy requests, draft_probs is not used in rejection sampling.

vllm/v1/worker/gpu_model_runner.py

@@ -2083,7 +2083,7 @@ class GPUModelRunner(LoRAModelRunnerMixin):
 
             if self.speculative_config and self.speculative_config.use_eagle():
                 assert isinstance(self.drafter, EagleProposer)
-                self.drafter.dummy_run(num_tokens)
+                self.drafter.dummy_run(num_tokens, attn_metadata)
 
         # This is necessary to avoid blocking DP.
         # For dummy runs, we typically skip EPLB since we don't have any real