[V1][Spec Decode] Implement Eagle Proposer [1/N]

vllm/config.py

@@ -2133,9 +2133,9 @@ def __post_init__(self):
 
                 # Replace hf_config for EAGLE draft_model
                 if self.method == "eagle":
-                    if self.enable_chunked_prefill:
-                        raise ValueError(
-                            "Chunked prefill and EAGLE are not compatible.")
+                    # if self.enable_chunked_prefill:
+                    #     raise ValueError(
+                    #         "Chunked prefill and EAGLE are not compatible.")
 
                     from vllm.transformers_utils.configs.eagle import (
                         EAGLEConfig)

vllm/engine/arg_utils.py

@@ -1647,7 +1647,9 @@
         if (self.speculative_model is not None
                 or self.num_speculative_tokens is not None):
             # This is supported but experimental (handled below).
-            if self.speculative_model in ("ngram", "[ngram]"):
+            supported_methods = ["ngram", "eagle"]
+            if any(method in self.speculative_model.lower()
+                   for method in supported_methods):
                 pass
             else:
                 _raise_or_fallback(feature_name="Speculative Decoding",

vllm/v1/spec_decode/eagle.py

@@ -0,0 +1,248 @@
+# SPDX-License-Identifier: Apache-2.0
+import torch
+import torch.nn as nn
+import triton
+import triton.language as tl
+
+from vllm.config import VllmConfig
+from vllm.forward_context import set_forward_context
+from vllm.v1.attention.backends.flash_attn import FlashAttentionMetadata
+from vllm.v1.sample.metadata import SamplingMetadata
+
+
+class EagleProposer:
+
+    def __init__(
+        self,
+        vllm_config: VllmConfig,
+        device: torch.device,
+    ):
+        self.vllm_config = vllm_config
+        self.num_speculative_tokens = (
+            vllm_config.speculative_config.num_speculative_tokens)
+        self.block_size = vllm_config.cache_config.block_size
+        self.arange = torch.arange(vllm_config.scheduler_config.max_num_seqs,
+                                   device=device)
+
+    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,
+        sampling_metadata: SamplingMetadata,
+    ) -> torch.Tensor:  # [batch_size, num_speculative_tokens]
+        num_tokens = target_token_ids.shape[0]
+        batch_size = next_token_ids.shape[0]
+        last_token_indices = cu_num_tokens[1:] - 1
+
+        input_ids = torch.empty_like(target_token_ids)
+        # Shift the input ids by one token.
+        # E.g., [a1, b1, b2, c1, c2, c3] -> [b1, b2, c1, c2, c3, c3]
+        input_ids[:-1] = target_token_ids[1:]
+        # Replace the last token with the next token.
+        # E.g., [b1, b2, c1, c2, c3, c3] -> [a2, b2, b3, c2, c3, c4]
+        input_ids[last_token_indices] = next_token_ids
+
+        seq_lens = target_positions[last_token_indices] + 1
+        # FIXME(woosuk): The below two ops cause synchronization. Optimize.
+        max_seq_len = seq_lens.max().item()
+        max_num_tokens = (cu_num_tokens[1:] - cu_num_tokens[:-1]).max().item()
+        attn_metadata = FlashAttentionMetadata(
+            num_actual_tokens=num_tokens,
+            max_query_len=max_num_tokens,
+            query_start_loc=cu_num_tokens,
+            max_seq_len=max_seq_len,
+            seq_lens=seq_lens,
+            block_table=block_table,
+            slot_mapping=target_slot_mapping,
+            # TODO(woosuk): Support cascade attention.
+            use_cascade=False,
+            common_prefix_len=0,
+            cu_prefix_query_lens=None,
+            prefix_kv_lens=None,
+            suffix_kv_lens=None,
+        )
+
+        with set_forward_context(attn_metadata, self.vllm_config):
+            hidden_states = self.model(
+                input_ids=input_ids,
+                hidden_states=target_hidden_states,
+                positions=target_positions,
+            )
+        sample_hidden_states = hidden_states[last_token_indices]
+        logits = self.model.compute_logits(sample_hidden_states, None)
+
+        # Sample the next token.
+        draft_token_ids = sample_token_ids(logits, sampling_metadata)
+
+        # Early exit if there is only one draft token to be generated.
+        if self.num_speculative_tokens == 1:
+            return draft_token_ids.view(-1, 1)
+
+        # Generate the remaining draft tokens.
+        draft_token_ids_list = [draft_token_ids]
+        positions = target_positions[last_token_indices]
+        hidden_states = sample_hidden_states
+        attn_metadata.num_actual_tokens = batch_size
+        attn_metadata.max_query_len = 1
+        attn_metadata.query_start_loc = self.arange[:batch_size]
+        for _ in range(self.num_speculative_tokens - 1):
+            # Update the inputs.
+            input_ids = draft_token_ids_list[-1]
+            positions += 1
+            attn_metadata.max_seq_len += 1
+            attn_metadata.seq_lens += 1
+            # Compute the slot mapping.
+            block_numbers = positions // self.block_size
+            block_ids = block_table.gather(dim=1,
+                                           index=block_numbers.view(-1, 1))
+            block_ids = block_ids.view(-1)
+            attn_metadata.slot_mapping = (block_ids * self.block_size +
+                                          positions % self.block_size)
+
+            # Run the model.
+            with set_forward_context(attn_metadata, self.vllm_config):
+                hidden_states = self.model(
+                    input_ids=input_ids,
+                    hidden_states=hidden_states,
+                    positions=positions,
+                )
+            logits = self.model.compute_logits(hidden_states, None)
+            draft_token_ids = sample_token_ids(logits, sampling_metadata)
+            draft_token_ids_list.append(draft_token_ids)
+
+        # [batch_size, num_speculative_tokens]
+        return torch.stack(draft_token_ids_list, dim=1)
+
+    @staticmethod
+    def prepare_inputs(
+        # [batch_size + 1]
+        cu_target_query_lens: torch.Tensor,
+        # [batch_size]
+        num_rejected_tokens: torch.Tensor,
+    ) -> tuple[torch.Tensor, torch.Tensor]:
+        # cu_target_query_lens: [0, a, a + b, a + b + c]
+        # num_rejected_tokens: [n1, n2, n3]
+        # num_tokens_per_req: [a - n1, b - n2, c - n3]
+        # cu_num_tokens: [0, a - n1, a + b - n1 - n2, a + b + c - n1 - n2 - n3]
+        # token_indices: [0, 1, ..., a - n1 - 1,
+        #                 a, a + 1, ..., a + b - n2 - 1,
+        #                 a + b, a + b + 1, ..., a + b + c - n3 - 1]
+
+        # [0, a, a + b, a + b + c] -> [a, b, c]
+        query_len_per_req = (cu_target_query_lens[1:] -
+                             cu_target_query_lens[:-1])
+        # [a, b, c] -> [a - n1, b - n2, c - n3]
+        num_tokens_per_req = query_len_per_req - num_rejected_tokens
+
+        cu_num_tokens = torch.empty_like(cu_target_query_lens)
+        torch.cumsum(num_tokens_per_req, dim=0, out=cu_num_tokens[1:])
+        cu_num_tokens[0] = 0
+
+        # FIXME(woosuk): Avoid synchronization.
+        num_tokens = cu_num_tokens[-1].item()
+        token_indices = torch.empty(
+            num_tokens,
+            dtype=torch.int32,
+            device=cu_num_tokens.device,
+        )
+
+        batch_size = num_rejected_tokens.shape[0]
+        BLOCK_SIZE = 1024
+        prepare_input_kernel[(batch_size, )](
+            token_indices,
+            cu_target_query_lens,
+            cu_num_tokens,
+            BLOCK_SIZE=BLOCK_SIZE,
+        )
+        return cu_num_tokens, token_indices
+
+    def load_model(self, target_model: nn.Module) -> None:
+        self.model = DummyEagleModel()
+        self.model.get_input_embeddings = target_model.get_input_embeddings
+        self.model.compute_logits = target_model.compute_logits
+
+
+# FIXME(woosuk): This is a dummy model for testing.
+# Remove this once we have a real model.
+class DummyEagleModel(nn.Module):
+
+    def __init__(self):
+        super().__init__()
+
+    def forward(
+        self,
+        input_ids: torch.Tensor,
+        hidden_states: torch.Tensor,
+        positions: torch.Tensor,
+    ) -> torch.Tensor:
+        input_embeddings = self.get_input_embeddings(input_ids)
+        return hidden_states + input_embeddings  # Dummy return.
+
+
+# TODO(woosuk): The logic here is duplicated with the main sampling code.
+# We should refactor this to reuse the same sampling implementation.
+def sample_token_ids(
+    logits: torch.Tensor,  # [batch_size, vocab_size]
+    sampling_metadata: SamplingMetadata,
+) -> torch.Tensor:  # [batch_size]
+    # NOTE(woosuk): We don't need to apply all the sampling parameters
+    # for generating the draft tokens.
+    if sampling_metadata.all_greedy:
+        # All greedy.
+        next_token_ids = logits.argmax(dim=-1)
+    else:
+        logits.div_(sampling_metadata.temperature.view(-1, 1))
+        probs = logits.softmax(dim=-1, dtype=torch.float32)
+
+        # TODO(woosuk): Consider seeds?
+        q = torch.empty_like(logits)
+        q.exponential_()
+        next_token_ids = probs.div_(q).argmax(dim=-1).view(-1)
+
+        if not sampling_metadata.all_random:
+            greedy_token_ids = logits.argmax(dim=-1)
+            next_token_ids = torch.where(
+                sampling_metadata.temperature == -1,
+                greedy_token_ids,
+                next_token_ids,
+            )
+    return next_token_ids
+
+
+@triton.jit
+def prepare_input_kernel(
+    out_ptr,
+    cu_query_lens_ptr,
+    cu_num_tokens_ptr,
+    BLOCK_SIZE: tl.constexpr,
+):
+    pid = tl.program_id(0)
+
+    # [start_pos, end_pos)
+    start_pos = tl.load(cu_num_tokens_ptr + pid)
+    end_pos = tl.load(cu_num_tokens_ptr + pid + 1)
+    num_tokens = end_pos - start_pos
+
+    index_start = tl.load(cu_query_lens_ptr + pid)
+    indices = index_start + tl.arange(0, BLOCK_SIZE)
+
+    num_blocks = tl.cdiv(num_tokens, BLOCK_SIZE)
+    for i in tl.range(num_blocks):
+        offset = i * BLOCK_SIZE + tl.arange(0, BLOCK_SIZE)
+        tl.store(
+            out_ptr + start_pos + offset,
+            indices,
+            mask=offset < num_tokens,
+        )

vllm/v1/worker/gpu_input_batch.py

@@ -39,9 +39,18 @@ class CachedRequestState:
 
     lora_request: Optional[LoRARequest] = None
 
+    def __post_init__(self):
+        self.num_prompt_tokens = len(self.prompt_token_ids)
+
     @property
     def num_tokens(self) -> int:
-        return len(self.prompt_token_ids) + len(self.output_token_ids)
+        return self.num_prompt_tokens + len(self.output_token_ids)
+
+    def get_token_id(self, idx: int) -> int:
+        if idx < self.num_prompt_tokens:
+            return self.prompt_token_ids[idx]
+        else:
+            return self.output_token_ids[idx - self.num_prompt_tokens]
 
 
 class InputBatch: