SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.10644 · NeurIPS · 2024

Improving Multi-candidate Speculative Decoding

Xiaofan Lu, Yixiao Zeng, Feiyang Ma, Zixu Yu, Marco Levorato

arXiv · PDF · Open in the Atlas

Code that ran

2 of 2 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 1 has no contract verdict; 1 failure of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2409.10644.json · badge: /api/ran/2409.10644.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
JackZeng0208/DynaSD marked official in Papers with Code
Papers with Code archive
2 ran · 0 no recorded run
FunctionStatusWhere it lives
TargetInitializedtMCSD Ran
ran on a synthesized input · constructed an object; no result was computed
JackZeng0208/DynaSD/inference/target_initialized_mcsd.py
code served (permissive licence) · get_code("aafc83632a107ba8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, TargetInitializedtMCSD: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from transformers.modeling_outputs import BaseModelOutputWithPast

torch.manual_seed(0)

# Create a minimal draft model
class MinimalDraftModel(nn.Module):
    def __init__(self, vocab_size=100, hidden_size=4, num_layers=2):
        super().__init__()
        self.model = MinimalModelCore(vocab_size, hidden_size, num_layers)
        self.lm_head = nn.Linear(hidden_size, vocab_size)
    
    def forward(self, input_ids, use_cache=True, past_key_values=None, return_dict=True, output_attentions=False, output_hidden_states=False, position_ids=None):
        return self.model(input_ids, use_cache, past_key_values, return_dict, output_attentions, output_hidden_states, position_ids)

# Create a minimal target model
class MinimalTargetModel(nn.Module):
    def __init__(self, vocab_size=100, hidden_size=4, num_layers=2):
        super().__init__()
        self.model = MinimalModelCore(vocab_size, hidden_size, num_layers)
        self.lm_head = nn.Linear(hidden_size, vocab_size)
    
    def forward(self, input_ids, use_cache=True, past_key_values=None, return_dict=True, output_attentions=False, output_hidden_states=False, position_ids=None):
        return self.model(input_ids, use_cache, past_key_values, return_dict, output_attentions, output_hidden_states, position_ids)

# Core model that supports the required interface
class MinimalModelCore(nn.Module):
    def __init__(self, vocab_size=100, hidden_size=4, num_layers=2):
        super().__init__()
        self.embedding = nn.Embedding(vocab_size, hidden_size)
        self.layers = nn.ModuleList([nn.Linear(hidden_size, hidden_size) for _ in range(num_layers)])
        self.hidden_size = hidden_size
    
    def get_input_embeddings(self):
        return self.embedding
    
    def forward(self, input_ids, use_cache=True, past_key_values=None, return_dict=True, output_attentions=False, output_hidden_states=False, position_ids=None):
        hidden_states = self.embedding(input_ids)
        
        for layer in self.layers:
            hidden_states = layer(hidden_states)
        
        # Create mock past_key_values (tuple of tuples for each layer)
        mock_past_key_values = tuple(
            (torch.zeros(1, 1, 2, self.hidden_size), torch.zeros(1, 1, 2, self.hidden_size))
            for _ in self.layers
        )
        
        output = BaseModelOutputWithPast(
            last_hidden_state=hidden_states,
            past_key_values=mock_past_key_values if use_cache else None,
            hidden_states=None,
            attentions=None
        )
        
        return output

# Create models
draft_model = MinimalDraftModel(vocab_size=100, hidden_size=4, num_layers=2)
target_model = MinimalTargetModel(vocab_size=100, hidden_size=4, num_layers=2)

# MCSD config: tree structure with [2, 2, 3] means 2 candidates at root, each has 2 children, each has 3 children
mcsd_config = (2, 2, 3)

# EOS token ID
eos_token_id = 2

ARGS = {
    'draft_model': draft_model,
    'target_model': target_model,
    'mcsd_config': mcsd_config,
    'eos_token_id': eos_token_id,
    'max_new_tokens': 200,
    'temperature': 0.0
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type TargetInitializedtMCSD — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a TargetInitializedtMCSD output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_tree_attn_self_mask Ran
ran; our contract draft was wrong, not the code
JackZeng0208/DynaSD/inference/target_initialized_mcsd.py
code served (permissive licence) · get_code("08d752372c58e83c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (20, 20) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct arguments for get_tree_attn_self_mask
# The function takes max_config as a Tuple[int] argument
# It represents the maximum configuration at each level of a tree structure

ARGS = {
    "max_config": (2, 3, 2),
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (20, 20) — shape (20, 20) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block

Also present in other papers' code

Of 2 samples lifted from this paper's linked repositories, 2 carry a function identity and 0 do not. Of the 2 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 2 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Speculative Decoding (SD) is a technique to accelerate the inference of Large Language Models (LLMs) by using a lower complexity draft model to propose candidate tokens verified by a larger target model. To further improve efficiency, Multi-Candidate Speculative Decoding (MCSD) improves upon this by sampling multiple candidate tokens from the draft model at each step and verifying them in parallel, thus increasing the chances of accepting a token and reducing generation time. Existing MCSD methods rely on the draft model to initialize the multi-candidate sequences and use static length and tree attention structure for draft generation. However, such an approach suffers from the draft and target model's output distribution differences, especially in a dynamic generation context. In this work, we introduce a new version of MCSD that includes a target model initialized multi-candidate generation, a dynamic sliced topology-aware causal mask for dynamic length adjustment, and decision models to optimize early stopping. We experimented with our method on Llama 2-7B and its variants and observed a maximum 27.5% speedup compared to our MCSD baseline across three benchmarks with Llama 2-7B as the target model and JackFram 68M as the draft model. Additionally, we evaluate the effects of using the target model initialized multi-candidate process with different draft models on output quality. Our original code is available on GitHub.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2409.10644")
get_code_for_paper("2409.10644")
have("2409.10644")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2409.10644.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2409.10644.svg)](https://syntology.ai/paper/2409.10644)

Connect an agent — have() is free.