[Model] MLPSpeculator speculative decoding support (#4947)

Signed-off-by: Thomas Parnell <tpa@zurich.ibm.com>

Co-authored-by: Thomas Parnell <tpa@zurich.ibm.com>
Co-authored-by: Nick Hill <nickhill@us.ibm.com>
Co-authored-by: Davis Wertheimer <Davis.Wertheimer@ibm.com>

examples/offline_inference_mlpspeculator.py

@@ -0,0 +1,59 @@
+import gc
+import time
+from typing import List
+
+from vllm import LLM, SamplingParams
+
+
+def time_generation(llm: LLM, prompts: List[str],
+                    sampling_params: SamplingParams):
+    # Generate texts from the prompts. The output is a list of RequestOutput
+    # objects that contain the prompt, generated text, and other information.
+    # Warmup first
+    llm.generate(prompts, sampling_params)
+    llm.generate(prompts, sampling_params)
+    start = time.time()
+    outputs = llm.generate(prompts, sampling_params)
+    end = time.time()
+    print((end - start) / sum([len(o.outputs[0].token_ids) for o in outputs]))
+    # Print the outputs.
+    for output in outputs:
+        generated_text = output.outputs[0].text
+        print(f"text: {generated_text!r}")
+
+
+if __name__ == "__main__":
+
+    template = (
+        "Below is an instruction that describes a task. Write a response "
+        "that appropriately completes the request.\n\n### Instruction:\n{}"
+        "\n\n### Response:\n")
+
+    # Sample prompts.
+    prompts = [
+        "Write about the president of the United States.",
+    ]
+    prompts = [template.format(prompt) for prompt in prompts]
+    # Create a sampling params object.
+    sampling_params = SamplingParams(temperature=0.0, max_tokens=200)
+
+    # Create an LLM without spec decoding
+    llm = LLM(model="meta-llama/Llama-2-13b-chat-hf")
+
+    print("Without speculation")
+    time_generation(llm, prompts, sampling_params)
+
+    del llm
+    gc.collect()
+
+    # Create an LLM with spec decoding
+    llm = LLM(
+        model="meta-llama/Llama-2-13b-chat-hf",
+        speculative_model="ibm-fms/llama-13b-accelerator",
+        # These are currently required for MLPSpeculator decoding
+        use_v2_block_manager=True,
+        enforce_eager=True,
+    )
+
+    print("With speculation")
+    time_generation(llm, prompts, sampling_params)

tests/spec_decode/test_spec_decode_worker.py

@@ -456,7 +456,9 @@ def test_k_equals_zero(k: int, batch_size: int):
     rejection_sampler.token_id_dtype = torch.int64
     metrics_collector = MagicMock(spec=AsyncMetricsCollector)
 
-    target_worker.execute_model.return_value = [MagicMock(spec=SamplerOutput)]
+    sampler_output = MagicMock(spec=SamplerOutput)
+    sampler_output.hidden_states = None
+    target_worker.execute_model.return_value = [sampler_output]
 
     draft_worker.device = 'cuda'
     target_worker.device = 'cuda'
@@ -497,7 +499,9 @@ def test_empty_input_batch(k: int, batch_size: int):
     rejection_sampler.token_id_dtype = torch.int64
     metrics_collector = MagicMock(spec=AsyncMetricsCollector)
 
-    target_worker.execute_model.return_value = [MagicMock(spec=SamplerOutput)]
+    sampler_output = MagicMock(spec=SamplerOutput)
+    sampler_output.hidden_states = None
+    target_worker.execute_model.return_value = [sampler_output]
 
     draft_worker.device = 'cuda'
     target_worker.device = 'cuda'

tests/spec_decode/test_utils.py

@@ -2,8 +2,8 @@ from unittest.mock import MagicMock
 
 import pytest
 
-from vllm.sequence import SequenceGroupMetadata
-from vllm.spec_decode.util import get_all_seq_ids, split_batch_by_proposal_len
+from vllm.sequence import SequenceGroupMetadata, get_all_seq_ids
+from vllm.spec_decode.util import split_batch_by_proposal_len
 
 
 def test_get_all_seq_ids():

vllm/config.py

@@ -230,7 +230,8 @@ class ModelConfig:
         self,
         parallel_config: "ParallelConfig",
     ) -> None:
-        total_num_attention_heads = self.hf_text_config.num_attention_heads
+        total_num_attention_heads = getattr(self.hf_text_config,
+                                            "num_attention_heads", 0)
         tensor_parallel_size = parallel_config.tensor_parallel_size
         if total_num_attention_heads % tensor_parallel_size != 0:
             raise ValueError(
@@ -238,7 +239,8 @@ class ModelConfig:
                 " must be divisible by tensor parallel size "
                 f"({tensor_parallel_size}).")
 
-        total_num_hidden_layers = self.hf_text_config.num_hidden_layers
+        total_num_hidden_layers = getattr(self.hf_text_config,
+                                          "num_hidden_layers", 0)
         pipeline_parallel_size = parallel_config.pipeline_parallel_size
         if total_num_hidden_layers % pipeline_parallel_size != 0:
             raise ValueError(
@@ -341,8 +343,8 @@ class ModelConfig:
 
     def get_num_attention_heads(self,
                                 parallel_config: "ParallelConfig") -> int:
-        return self.hf_text_config.num_attention_heads // \
-            parallel_config.tensor_parallel_size
+        num_heads = getattr(self.hf_text_config, "num_attention_heads", 0)
+        return num_heads // parallel_config.tensor_parallel_size
 
     def get_num_layers(self, parallel_config: "ParallelConfig") -> int:
         total_num_hidden_layers = self.hf_text_config.num_hidden_layers
@@ -818,7 +820,8 @@ class SpeculativeConfig:
             speculative_model (Optional[str]): The name of the speculative
                 model, if provided.
             num_speculative_tokens (Optional[int]): The number of speculative
-                tokens, if provided.
+                tokens, if provided. Will default to the number in the draft
+                model config if present, otherwise is required.
             speculative_max_model_len (Optional[int]): The maximum model len of
                 the speculative model. Used when testing the ability to skip
                 speculation for some sequences.
@@ -841,24 +844,18 @@ class SpeculativeConfig:
                 the necessary conditions are met, else None.
         """
 
-        if speculative_model is None and num_speculative_tokens is None:
+        if speculative_model is None:
+            if num_speculative_tokens is not None:
+                raise ValueError("num_speculative_tokens was provided without "
+                                 "speculative_model.")
             return None
 
-        if speculative_model is not None and num_speculative_tokens is None:
-            raise ValueError(
-                "Expected both speculative_model and "
-                "num_speculative_tokens to be provided, but found "
-                f"{speculative_model=} and {num_speculative_tokens=}.")
-
         if (speculative_disable_by_batch_size is not None
                 and speculative_disable_by_batch_size < 2):
             raise ValueError("Expect the batch size threshold of disabling "
                              "speculative decoding is > 1, but got "
                              f"{speculative_disable_by_batch_size=}")
 
-        assert (speculative_model is not None
-                and num_speculative_tokens is not None)
-
         if enable_chunked_prefill:
             raise ValueError(
                 "Speculative decoding and chunked prefill are "
@@ -912,6 +909,27 @@ class SpeculativeConfig:
                 max_logprobs=target_model_config.max_logprobs,
             )
 
+            if (draft_model_config.hf_config.model_type == "mlp_speculator"
+                    and target_parallel_config.world_size != 1):
+                # MLPSpeculator TP support will be added very soon
+                raise ValueError(
+                    "Speculative decoding with mlp_speculator models does not "
+                    "yet support distributed inferencing (TP > 1).")
+
+            n_predict = getattr(draft_model_config.hf_config, "n_predict",
+                                None)
+            if n_predict is not None:
+                if num_speculative_tokens is None:
+                    # Default to max value defined in draft model config.
+                    num_speculative_tokens = n_predict
+                elif num_speculative_tokens > n_predict:
+                    # Verify provided value doesn't exceed the maximum
+                    # supported by the draft model.
+                    raise ValueError(
+                        "Expected both speculative_model and "
+                        "num_speculative_tokens to be provided, but found "
+                        f"{speculative_model=} and {num_speculative_tokens=}.")
+
             draft_model_config.max_model_len = (
                 SpeculativeConfig._maybe_override_draft_max_model_len(
                     speculative_max_model_len,
@@ -923,6 +941,12 @@ class SpeculativeConfig:
                 SpeculativeConfig.create_draft_parallel_config(
                     target_parallel_config))
 
+        if num_speculative_tokens is None:
+            raise ValueError(
+                "num_speculative_tokens must be provided with "
+                "speculative_model unless the draft model config contains an "
+                "n_predict parameter.")
+
         return SpeculativeConfig(
             draft_model_config,
             draft_parallel_config,

vllm/model_executor/models/__init__.py

@@ -60,6 +60,7 @@ _GENERATION_MODELS = {
     "ArcticForCausalLM": ("arctic", "ArcticForCausalLM"),
     "XverseForCausalLM": ("xverse", "XverseForCausalLM"),
     "Phi3SmallForCausalLM": ("phi3_small", "Phi3SmallForCausalLM"),
+    "MLPSpeculatorPreTrainedModel": ("mlp_speculator", "MLPSpeculator"),
 }
 
 _EMBEDDING_MODELS = {

vllm/model_executor/models/mlp_speculator.py

@@ -0,0 +1,143 @@
+import math
+from typing import Iterable, List, Tuple
+
+import torch
+import torch.nn as nn
+
+from vllm.model_executor import SamplingMetadata
+from vllm.model_executor.layers.logits_processor import LogitsProcessor
+from vllm.model_executor.layers.sampler import Sampler
+from vllm.model_executor.layers.vocab_parallel_embedding import (
+    VocabParallelEmbedding)
+from vllm.model_executor.model_loader.weight_utils import default_weight_loader
+from vllm.sequence import SamplerOutput
+
+
+class MLPSpeculatorLayerNorm(nn.Module):
+    """
+    A L2 normalization implementation
+    ...
+    Args
+    ----
+    normalized_shape : int
+        Dimensionality of input data (size of final tensor axis)
+    eps : float
+        Safety term to prevent division by zero. Make sure the chosen value
+         fits in the range of your encoding scheme
+         (i.e. fp16 requires eps >= 6e-8).
+    """
+
+    def __init__(
+        self,
+        normalized_shape,
+        eps=1e-06,
+    ):
+        super(MLPSpeculatorLayerNorm, self).__init__()
+        self.weight = nn.Parameter(torch.empty(normalized_shape))
+        self.bias = nn.Parameter(torch.empty(normalized_shape))
+        self.eps = eps
+
+    def forward(self, x):
+        xf = x
+        xf = xf * torch.rsqrt(xf.pow(2).mean(-1, keepdim=True) + self.eps)
+        x = xf.type_as(x)
+        x = self.weight * x
+        x = x + self.bias
+        return x
+
+
+class MLPSpeculator(nn.Module):
+
+    def __init__(self, config, **kwargs) -> None:
+        super().__init__()
+        self.n_predict = config.n_predict
+        self.vocab_size = config.vocab_size
+        self.emb_dim = config.emb_dim
+        self.inner_dim = config.inner_dim if config.inner_dim != 0 \
+            else config.emb_dim
+
+        self.max_speculative_tokens = getattr(config, "max_speculative_tokens",
+                                              self.n_predict)
+
+        self.emb = nn.ModuleList([
+            VocabParallelEmbedding(config.vocab_size,
+                                   self.inner_dim,
+                                   org_num_embeddings=config.vocab_size)
+            for _ in range(self.max_speculative_tokens)
+        ])
+
+        self.proj = nn.ModuleList([
+            nn.Linear((self.emb_dim if i == 0 else self.inner_dim),
+                      self.inner_dim,
+                      bias=False) for i in range(self.max_speculative_tokens)
+        ])
+
+        self.head = nn.ModuleList([
+            nn.Linear(self.inner_dim, self.vocab_size, bias=False)
+            for _ in range(self.max_speculative_tokens)
+        ])
+        self.ln = nn.ModuleList([
+            MLPSpeculatorLayerNorm(self.inner_dim)
+            for _ in range(self.max_speculative_tokens)
+        ])
+
+        self.state_weight = 0.5**(0.5 / config.n_predict)
+        self.emb_weight = math.sqrt(
+            (1 - self.state_weight**2) * (self.inner_dim / 2))
+        self.activation = nn.GELU()
+        self.config = config
+        self.logits_processor = LogitsProcessor(config.vocab_size,
+                                                config.vocab_size, 1.0)
+        self.sampler = Sampler()
+
+    def generate_proposals(
+        self,
+        input_ids: torch.Tensor,
+        previous_hidden_states: torch.Tensor,
+        num_predict_tokens: int,
+        sampling_metadata: SamplingMetadata,
+    ) -> List[SamplerOutput]:
+        if num_predict_tokens > self.max_speculative_tokens:
+            raise ValueError(f"Max speculative tokens for model is "
+                             f"{self.max_speculative_tokens}, but "
+                             f"{num_predict_tokens} were requested")
+
+        # b x 1 x d
+        previous_hidden_states = previous_hidden_states.unsqueeze(1)
+
+        # b x 1
+        last_tokens = input_ids.unsqueeze(1)
+
+        next_tokens = []
+
+        for head_index in range(num_predict_tokens):
+
+            # Project and predict
+            z = self.emb[head_index](last_tokens)  # b k d
+            states = self.proj[head_index](previous_hidden_states)
+
+            # Weighted add of state_weight*state and emb_weight*z
+            # Let subsequent LN take care of denominator
+            # state_weight is close to 1, so shouldn't be any precision issues
+            states.add_(z, alpha=self.emb_weight / self.state_weight)
+
+            states = self.activation(self.ln[head_index](states))  # b k d
+            # TODO: not yet supporting top_k_tokens_per_head
+            previous_hidden_states = states
+
+            logits = self.logits_processor(self.head[head_index].weight,
+                                           states, sampling_metadata)
+
+            output = self.sampler(logits.flatten(0, 1), sampling_metadata)
+            last_tokens = output.sampled_token_ids
+            next_tokens.append(output)
+
+        return next_tokens
+
+    def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
+        params_dict = dict(self.named_parameters())
+        for name, loaded_weight in weights:
+            param = params_dict[name.replace("speculator.", "")]
+            weight_loader = getattr(param, "weight_loader",
+                                    default_weight_loader)
+            weight_loader(param, loaded_weight)

vllm/sequence.py

@@ -794,6 +794,9 @@ class SamplerOutput:
     # Spec decode metrics populated by workers.
     spec_decode_worker_metrics: Optional["SpecDecodeWorkerMetrics"] = None
 
+    # Optional last hidden states from the model.
+    hidden_states: Optional[torch.Tensor] = None
+
     def __getitem__(self, idx: int):
         return self.outputs[idx]
 
@@ -842,6 +845,46 @@ class PoolerOutput:
                           self.__class__) and self.outputs == other.outputs
 
 
+def get_all_seq_ids(
+        seq_group_metadata_list: List[SequenceGroupMetadata]) -> List[int]:
+    """Given a list of SequenceGroupMetadata, create a list of all
+    sequence ids.
+    """
+    return [seq_id for sg in seq_group_metadata_list for seq_id in sg.seq_data]
+
+
+class HiddenStates:
+    """Hidden states corresponding to in-progress sequences.
+    Used in speculative decoding to pass hidden states from
+    the target model to the proposer model in the subsequent step.
+
+    seq_ids are the sequence ids of each entry of the batch
+    dimension of the hidden_states tensor"""
+
+    def __init__(self, seq_group_metadata_list: List[SequenceGroupMetadata],
+                 hidden_states: torch.Tensor):
+        assert len(seq_group_metadata_list) == len(hidden_states)
+        self.seq_ids: List[int] = get_all_seq_ids(seq_group_metadata_list)
+        self.hidden_states: torch.Tensor = hidden_states
+
+    def update(self, seq_group_metadata_list: List[SequenceGroupMetadata],
+               hidden_states: torch.Tensor) -> None:
+        """Update hidden states from target model invocation."""
+        assert len(seq_group_metadata_list) == len(hidden_states)
+        self.seq_ids.extend(get_all_seq_ids(seq_group_metadata_list))
+        self.hidden_states = torch.cat([self.hidden_states, hidden_states])
+
+    def prune(self,
+              seq_group_metadata_list: List[SequenceGroupMetadata]) -> None:
+        """Prune to provided list of sequence ids."""
+        seq_ids = get_all_seq_ids(seq_group_metadata_list)
+        if seq_ids != self.seq_ids:
+            # Batch contents changed - prune removed sequences.
+            index = [self.seq_ids.index(seq_id) for seq_id in seq_ids]
+            self.hidden_states = self.hidden_states[index]
+            self.seq_ids = seq_ids
+
+
 @dataclass
 class ExecuteModelRequest:
     """The model execution request."""
@@ -857,6 +900,8 @@ class ExecuteModelRequest:
     num_lookahead_slots: int = 0
     # The number of requests in the running queue.
     running_queue_size: int = 0
+    # Optional hidden states from prior step.
+    previous_hidden_states: Optional[HiddenStates] = None
 
     def clone(
         self, seq_group_metadata_list: List[SequenceGroupMetadata]
@@ -869,4 +914,5 @@ class ExecuteModelRequest:
             blocks_to_copy=self.blocks_to_copy.copy(),
             num_lookahead_slots=self.num_lookahead_slots,
             running_queue_size=self.running_queue_size,
+            previous_hidden_states=self.previous_hidden_states,
         )

vllm/spec_decode/batch_expansion.py

@@ -4,11 +4,10 @@ from typing import Iterator, List, Tuple
 import torch
 
 from vllm.sequence import (ExecuteModelRequest, SamplerOutput, SequenceData,
-                           SequenceGroupMetadata)
+                           SequenceGroupMetadata, get_all_seq_ids)
 from vllm.spec_decode.interfaces import (SpeculativeProposals,
                                          SpeculativeScorer, SpeculativeScores)
-from vllm.spec_decode.util import (get_all_seq_ids, nvtx_range,
-                                   sampler_output_to_torch,
+from vllm.spec_decode.util import (nvtx_range, sampler_output_to_torch,
                                    split_batch_by_proposal_len)
 from vllm.worker.worker_base import WorkerBase
 
@@ -98,6 +97,7 @@ class BatchExpansionTop1Scorer(SpeculativeScorer):
             probs=all_probs,
             token_ids=all_tokens,
             logprobs=spec_logprobs,
+            hidden_states=target_sampler_output.hidden_states,
         )
 
     def _expand_batch(

vllm/spec_decode/interfaces.py

@@ -1,5 +1,6 @@
 from abc import ABC, abstractmethod
 from dataclasses import dataclass
+from typing import Optional
 
 import torch
 
@@ -46,6 +47,9 @@ class SpeculativeScores:
     # tokens and also non-speculative normal decoding.
     token_ids: torch.Tensor
 
+    # Optional last hidden states from the scoring model.
+    hidden_states: Optional[torch.Tensor] = None
+
     def __repr__(self):
         return (f"SpeculativeScores("
                 f"probs={self.probs.shape}, "

vllm/spec_decode/mlp_speculator_worker.py

@@ -0,0 +1,87 @@
+from typing import List, Optional, Tuple
+
+import torch
+
+from vllm.model_executor import SamplingMetadata
+from vllm.sequence import (ExecuteModelRequest, SamplerOutput,
+                           SequenceGroupMetadata)
+from vllm.spec_decode.multi_step_worker import MultiStepWorker
+from vllm.spec_decode.proposer_worker_base import NonLLMProposerWorkerBase
+from vllm.worker.model_runner import ModelInput
+
+
+class MLPSpeculatorWorker(NonLLMProposerWorkerBase, MultiStepWorker):
+    """Worker for MLPSpeculator models.
+
+    Not currently compatible with LoRA or chunked prefill.
+    """
+
+    @torch.inference_mode()
+    def sampler_output(
+        self,
+        execute_model_req: ExecuteModelRequest,
+        sample_len: int,
+    ) -> Tuple[List[SamplerOutput], bool]:
+        """Run the model forward pass to generate sample_len future tokens.
+        Returns the list of sampler output, one per layer, along with indicator
+        of whether torch tensor in sampler output need to be transposed in
+        latter sampler_output_to_torch logic.
+
+        For mlp spec worker, this indicator shall be True.
+        """
+        self._raise_if_unsupported(execute_model_req)
+
+        seq_group_metadata_list = execute_model_req.seq_group_metadata_list
+
+        (input_tokens, seq_lens,
+         query_lens) = self._prepare_input_tensors(seq_group_metadata_list)
+
+        sampling_metadata = SamplingMetadata.prepare(
+            seq_group_metadata_list, seq_lens, query_lens, self.device,
+            self.model_runner.pin_memory)
+
+        model_outputs = self.model_runner.model.generate_proposals(
+            input_ids=input_tokens,
+            previous_hidden_states=execute_model_req.previous_hidden_states.
+            hidden_states,
+            num_predict_tokens=sample_len,
+            sampling_metadata=sampling_metadata)
+
+        assert len(model_outputs) == sample_len
+
+        return model_outputs, True
+
+    def _prepare_input_tensors(
+        self,
+        seq_group_metadata_list: Optional[List[SequenceGroupMetadata]],
+    ) -> Tuple[torch.Tensor, List[int], List[int]]:
+        if not seq_group_metadata_list:
+            return ModelInput.empty(self.device)
+
+        input_tokens: List[int] = []
+        seq_lens: List[int] = []
+        query_lens: List[int] = []
+
+        for seq_group_metadata in seq_group_metadata_list:
+            is_prompt = seq_group_metadata.is_prompt
+
+            for seq_data in seq_group_metadata.seq_data.values():
+                seq_data_len = seq_data.get_len()
+                if is_prompt:
+                    context_len = seq_data.get_num_computed_tokens()
+                    seq_len = min(
+                        seq_data_len,
+                        context_len + seq_group_metadata.token_chunk_size)
+                    tokens = seq_data.get_token_ids()[context_len:seq_len]
+                    seq_lens.append(seq_len)
+                    input_tokens.extend(tokens)
+                    query_lens.append(seq_len - context_len)
+                else:
+                    seq_lens.append(seq_data_len)
+                    input_tokens.append(seq_data.get_last_token_id())
+                    query_lens.append(1)
+
+        input_tokens_tensor = torch.tensor(input_tokens,
+                                           dtype=torch.long,
+                                           device=self.device)
+        return input_tokens_tensor, seq_lens, query_lens

vllm/spec_decode/spec_decode_worker.py

@@ -8,16 +8,18 @@ from vllm.distributed.communication_op import broadcast_tensor_dict
 from vllm.logger import init_logger
 from vllm.model_executor.layers.rejection_sampler import RejectionSampler
 from vllm.sequence import (CompletionSequenceGroupOutput, ExecuteModelRequest,
-                           SamplerOutput, SequenceGroupMetadata)
+                           HiddenStates, SamplerOutput, SequenceGroupMetadata,
+                           get_all_seq_ids)
 from vllm.spec_decode.batch_expansion import BatchExpansionTop1Scorer
 from vllm.spec_decode.interfaces import (SpeculativeProposals,
                                          SpeculativeScorer, SpeculativeScores)
 from vllm.spec_decode.metrics import AsyncMetricsCollector
+from vllm.spec_decode.mlp_speculator_worker import MLPSpeculatorWorker
 from vllm.spec_decode.multi_step_worker import MultiStepWorker
 from vllm.spec_decode.ngram_worker import NGramWorker
 from vllm.spec_decode.proposer_worker_base import ProposerWorkerBase
 from vllm.spec_decode.util import (create_sequence_group_output,
-                                   get_all_num_logprobs, get_all_seq_ids,
+                                   get_all_num_logprobs,
                                    get_sampled_token_logprobs, nvtx_range,
                                    split_batch_by_proposal_len)
 from vllm.worker.worker import Worker
@@ -104,6 +106,10 @@ class SpecDecodeWorker(LoraNotSupportedWorkerBase):
             proposer_worker = NGramWorker(**draft_worker_kwargs)
             proposer_worker.set_ngram_window_size(ngram_prompt_lookup_min,
                                                   ngram_prompt_lookup_max)
+        elif draft_worker_kwargs[
+                "model_config"].hf_config.model_type == "mlp_speculator":
+            proposer_worker = MLPSpeculatorWorker(**draft_worker_kwargs)
+            disable_bonus_tokens = False
         else:
             proposer_worker = MultiStepWorker(**draft_worker_kwargs)
 
@@ -155,6 +161,10 @@ class SpecDecodeWorker(LoraNotSupportedWorkerBase):
         # Lazy initiazliation.
         self.scorer: SpeculativeScorer
 
+        # Hidden states from target model to pass to proposer
+        # in the subsequent step.
+        self.previous_hidden_states: Optional[HiddenStates] = None
+
     def init_device(self) -> None:
         """Initialize both scorer and proposer models.
         """
@@ -337,6 +347,16 @@ class SpecDecodeWorker(LoraNotSupportedWorkerBase):
         assert len(sampler_output) == 1
         sampler_output = sampler_output[0]
 
+        # Store hidden states from target model execution.
+        hidden_states = sampler_output.hidden_states
+        if hidden_states is not None:
+            if self.previous_hidden_states is None:
+                self.previous_hidden_states = HiddenStates(
+                    execute_model_req.seq_group_metadata_list, hidden_states)
+            else:
+                self.previous_hidden_states.update(
+                    execute_model_req.seq_group_metadata_list, hidden_states)
+
         # Clear device tensors from sampler output. This reduces communication
         # overhead when the engine runs in a different process than the workers.
         sampler_output.probs = None
@@ -383,6 +403,10 @@ class SpecDecodeWorker(LoraNotSupportedWorkerBase):
         """
         assert num_lookahead_slots == execute_model_req.num_lookahead_slots
 
+        # Pass last hidden states from target model to proposer
+        execute_model_req.previous_hidden_states = self.previous_hidden_states
+        self.previous_hidden_states = None
+
         # Generate proposals using draft worker.
         proposals = self.proposer_worker.get_spec_proposals(execute_model_req)
 
@@ -466,6 +490,20 @@ class SpecDecodeWorker(LoraNotSupportedWorkerBase):
         # metadata.
         accepted_token_ids[original_indices] = accepted_token_ids.clone()
 
+        hidden_states = proposal_scores.hidden_states
+        if hidden_states is not None:
+            # Contract hidden states based on accepted tokens
+            hs_size = hidden_states.shape[1]
+            hidden_states = hidden_states.reshape(-1, max_proposal_len + 1,
+                                                  hs_size)
+            accepted_index = accepted_token_ids + 1  # Convert -1 to 0
+            accepted_index = accepted_index.count_nonzero(dim=1).add_(-1)
+            index = accepted_index[:, None, None].expand(-1, 1, hs_size)
+            hidden_states = hidden_states.gather(1, index).squeeze(1)  # b x d
+            # Store hidden states from target model for subsequent decode step
+            self.previous_hidden_states = HiddenStates(seq_group_metadata_list,
+                                                       hidden_states)
+
         return accepted_token_ids, logprobs
 
     def _create_output_sampler_list(

vllm/spec_decode/top1_proposer.py

@@ -65,9 +65,13 @@ class Top1Proposer(SpeculativeProposer):
             # token_ids is like [batch] format in proposal_len size list,
             # while if it is false, the format would be [proposal_len]
             # in batch size list
+            hidden_states = execute_model_req.previous_hidden_states
+            if hidden_states is not None:
+                hidden_states.prune(nonzero_proposal_len_seqs)
             nonzero_execute_model_req = ExecuteModelRequest(
                 seq_group_metadata_list=nonzero_proposal_len_seqs,
                 num_lookahead_slots=proposal_len,
+                previous_hidden_states=hidden_states,
             )
             maybe_sampler_output, transposed = self._worker.sampler_output(
                 execute_model_req=nonzero_execute_model_req,

vllm/spec_decode/util.py

@@ -10,14 +10,6 @@ from vllm.sequence import (CompletionSequenceGroupOutput, Logprob,
 SeqId = int
 
 
-def get_all_seq_ids(
-        seq_group_metadata_list: List[SequenceGroupMetadata]) -> List[SeqId]:
-    """Given a list of SequenceGroupMetadata, create a list of all
-    sequence ids.
-    """
-    return [seq_id for sg in seq_group_metadata_list for seq_id in sg.seq_data]
-
-
 def get_all_num_logprobs(
         seq_group_metadata_list: List[SequenceGroupMetadata]) -> List[int]:
     """Given a list of SequenceGroupMetadata, create a list of all num_logprobs.

vllm/transformers_utils/config.py

@@ -1,3 +1,4 @@
+import contextlib
 from typing import Dict, Optional, Type
 
 from transformers import PretrainedConfig
@@ -5,7 +6,13 @@ from transformers import PretrainedConfig
 from vllm.envs import VLLM_USE_MODELSCOPE
 from vllm.logger import init_logger
 from vllm.transformers_utils.configs import (ChatGLMConfig, DbrxConfig,
-                                             JAISConfig, MPTConfig, RWConfig)
+                                             JAISConfig, MLPSpeculatorConfig,
+                                             MPTConfig, RWConfig)
+
+if VLLM_USE_MODELSCOPE:
+    from modelscope import AutoConfig
+else:
+    from transformers import AutoConfig
 
 logger = init_logger(__name__)
 
@@ -16,8 +23,13 @@ _CONFIG_REGISTRY: Dict[str, Type[PretrainedConfig]] = {
     "RefinedWeb": RWConfig,  # For tiiuae/falcon-40b(-instruct)
     "RefinedWebModel": RWConfig,  # For tiiuae/falcon-7b(-instruct)
     "jais": JAISConfig,
+    "mlp_speculator": MLPSpeculatorConfig,
 }
 
+for name, cls in _CONFIG_REGISTRY.items():
+    with contextlib.suppress(ValueError):
+        AutoConfig.register(name, cls)
+
 
 def get_config(model: str,
                trust_remote_code: bool,
@@ -26,10 +38,6 @@ def get_config(model: str,
                rope_scaling: Optional[dict] = None,
                rope_theta: Optional[float] = None) -> PretrainedConfig:
     try:
-        if VLLM_USE_MODELSCOPE:
-            from modelscope import AutoConfig
-        else:
-            from transformers import AutoConfig
         config = AutoConfig.from_pretrained(
             model,
             trust_remote_code=trust_remote_code,

vllm/transformers_utils/configs/__init__.py

@@ -5,6 +5,7 @@ from vllm.transformers_utils.configs.dbrx import DbrxConfig
 # `FalconConfig` class from the official HuggingFace transformers library.
 from vllm.transformers_utils.configs.falcon import RWConfig
 from vllm.transformers_utils.configs.jais import JAISConfig
+from vllm.transformers_utils.configs.mlp_speculator import MLPSpeculatorConfig
 from vllm.transformers_utils.configs.mpt import MPTConfig
 
 __all__ = [
@@ -13,4 +14,5 @@ __all__ = [
     "MPTConfig",
     "RWConfig",
     "JAISConfig",
+    "MLPSpeculatorConfig",
 ]

vllm/transformers_utils/configs/mlp_speculator.py

@@ -0,0 +1,50 @@
+from typing import List, Optional
+
+from transformers import PretrainedConfig
+
+
+class MLPSpeculatorConfig(PretrainedConfig):
+    model_type = "mlp_speculator"
+
+    attribute_map = {
+        "hidden_size": "emb_dim",
+    }
+
+    def __init__(self,
+                 vocab_size: int = 32000,
+                 emb_dim: int = 4096,
+                 inner_dim: int = 0,
+                 n_predict: int = 3,
+                 top_k_tokens_per_head: Optional[List[int]] = None,
+                 n_candidates: int = 5,
+                 **kwargs):
+        """
+        Initialize an MLPSpeculatorConfig
+
+        Args:
+            vocab_size: int
+                the model vocab size
+            emb_dim: int
+                the model embedding dimension
+            inner_dim: int
+                the inner dimension of the model. If 0, will be the emb_dim.
+            n_predict: int
+                the number of lookaheads for the speculator
+            top_k_tokens_per_head: List[int]
+                Number of tokens to consider from each head when forming the
+                candidate tree.
+                For each candidate branch in the tree, head n produces topk[n]
+                additional sub-branches.
+            n_candidates: int
+                number of child candidates to create per sequence
+        """
+        if top_k_tokens_per_head is None:
+            top_k_tokens_per_head = [5, 4, 3]
+        assert len(top_k_tokens_per_head) == n_predict
+        self.vocab_size = vocab_size
+        self.emb_dim = emb_dim
+        self.inner_dim = inner_dim
+        self.n_predict = n_predict
+        self.top_k_tokens_per_head = top_k_tokens_per_head
+        self.n_candidates = n_candidates
+        super().__init__(**kwargs)

vllm/worker/model_runner.py

@@ -86,6 +86,7 @@ class ModelRunner:
         kv_cache_dtype: Optional[str] = "auto",
         is_driver_worker: bool = False,
         vision_language_config: Optional[VisionLanguageConfig] = None,
+        return_hidden_states: bool = False,
     ):
         self.model_config = model_config
         self.parallel_config = parallel_config
@@ -96,6 +97,7 @@ class ModelRunner:
         self.load_config = load_config
         self.is_driver_worker = is_driver_worker
         self.vision_language_config = vision_language_config
+        self.return_hidden_states = return_hidden_states
 
         self.device = self.device_config.device
         self.pin_memory = is_pin_memory_available()
@@ -116,15 +118,17 @@ class ModelRunner:
         self.graph_block_tables = np.zeros(
             (max(_BATCH_SIZES_TO_CAPTURE), self.get_max_block_per_batch()),
             dtype=np.int32)
+        num_attn_heads = self.model_config.get_num_attention_heads(
+            self.parallel_config)
         self.attn_backend = get_attn_backend(
-            self.model_config.get_num_attention_heads(self.parallel_config),
+            num_attn_heads,
             self.model_config.get_head_size(),
             self.model_config.get_num_kv_heads(self.parallel_config),
             self.model_config.get_sliding_window(),
             self.model_config.dtype,
             self.kv_cache_dtype,
             self.block_size,
-        )
+        ) if num_attn_heads else None
 
         # Create processor for multi-modal data
         if self.vision_language_config is not None:
@@ -762,11 +766,19 @@ class ModelRunner:
             return None
 
         # Sample the next token.
-        output = self.model.sample(
+        output: SamplerOutput = self.model.sample(
             logits=logits,
             sampling_metadata=sampling_metadata,
         )
 
+        if self.return_hidden_states:
+            # we only need to pass hidden states of most recent token
+            assert seq_group_metadata_list is not None
+            if seq_group_metadata_list[0].is_prompt:
+                hidden_states = hidden_states.index_select(
+                    0, sampling_metadata.selected_token_indices)
+            output.hidden_states = hidden_states
+
         return output
 
     @torch.inference_mode()

vllm/worker/worker.py

@@ -70,6 +70,14 @@ class Worker(WorkerBase):
             assert not self.lora_config, (
                 "To be tested: vision language model with LoRA settings.")
 
+        # Return hidden states from target model if the draft model is an
+        # mlp_speculator
+        speculative_args = {} if speculative_config is None \
+            or (speculative_config.draft_model_config.model ==
+                model_config.model) \
+              or (speculative_config.draft_model_config.hf_config.model_type !=
+                  "mlp_speculator") else {"return_hidden_states": True}
+
         ModelRunnerClass = (EmbeddingModelRunner if
                             self.model_config.embedding_mode else ModelRunner)
         self.model_runner = ModelRunnerClass(
@@ -83,6 +91,7 @@ class Worker(WorkerBase):
             kv_cache_dtype=self.cache_config.cache_dtype,
             is_driver_worker=is_driver_worker,
             vision_language_config=vision_language_config,
+            **speculative_args,
         )
         # Uninitialized cache engine. Will be initialized by
         # initialize_cache.