SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.01219 · 2026

Mixture-of-Top-k Attention: Efficient Attention via Scalable Fast Weights

Wei He, Zhiyuan Huang, Xianghan Meng, Chun-Guang Li, Qishuai Wen

arXiv · PDF · Open in the Atlas

Code that ran

2 of 3 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 2 failures of our instrument; 1 no recorded run), measured 2026-11-22; 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/2602.01219.json · badge: /api/ran/2602.01219.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
QishuaiWen/MiTA canonical (classifier label)
classifier basis: census
2 ran · 1 no recorded run
FunctionStatusWhere it lives
build_transform Ran
ran; our contract draft was wrong, not the code
QishuaiWen/MiTA/MiTA-DeiT/datasets.py
code served (permissive licence) · get_code("ebeb7b6e20a434cf")
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 Compose. 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 numpy as np
from torchvision import transforms
from timm.data.constants import IMAGENET_DEFAULT_MEAN, IMAGENET_DEFAULT_STD

torch.manual_seed(0)

class ArgsConfig:
    def __init__(self):
        self.input_size = 224
        self.color_jitter = 0.4
        self.aa = 'rand-m9-mstd0.5-inc1'
        self.train_interpolation = 'bicubic'
        self.reprob = 0.25
        self.remode = 'pixel'
        self.recount = 1
        self.eval_crop_ratio = 0.875

args = ArgsConfig()

ARGS = {
    'is_train': True,
    'args': args
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, timm, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Compose — a Compose output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Compose 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
conv3x3 Ran
ran; our contract draft was wrong, not the code
QishuaiWen/MiTA/MiTA-DeiT/patchconvnet_models.py
code served (permissive licence) · get_code("ba6aa5f07daca9cd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned Sequential. 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

torch.manual_seed(0)

ARGS = {
    'in_planes': 8,
    'out_planes': 16,
    'stride': 2
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 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
build_dataset Not yet run
unverified (harvested, no recorded run)
QishuaiWen/MiTA/MiTA-DeiT/datasets.py
code served (permissive licence) · get_code("d58967bba23da4e6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 3 samples lifted from this paper's linked repositories, 3 carry a function identity and 0 do not. Of the 3 distinct functions with an identity, 3 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 0 the same month, 0 later) and 0 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.

FunctionAlso present inEarliest-datedOthers named
build_dataset 25 other papers
25 dated earlier
2104.05707 (2021-04)
LocalViT: Analyzing Locality in Vision Transformers
2106.00515 (2021-06)
KVT: k-NN Attention for Boosting Vision Transformers
2106.01401 (2021-06)
Container: Context Aggregation Network
and 22 more not named here
build_transform 6 other papers
6 dated earlier
2310.01812 (2023-10)
PPT: Token Pruning and Pooling for Efficient Vision Transformers
2312.00752 (2023-12)
Mamba: Linear-Time Sequence Modeling with Selective State Spaces
2403.01590 (2024-03)
The Hidden Attention of Mamba Models
and 3 more not named here
conv3x3 4 other papers
4 dated earlier
2112.13692 (2021-12)
Augmenting Convolutional networks with attention-based aggregation
Yu_LaPE_Layer-adaptive_Position_Embedding_for_Vision_Transformers_with_Independent_Layer_ICCV_2023_paper (2023 (year only))
LaPE: Layer-adaptive Position Embedding for Vision Transformers with Independent Layer Normalization
2310.01812 (2023-10)
PPT: Token Pruning and Pooling for Efficient Vision Transformers
and 1 more not named here

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

The vanilla self-attention mechanism in Transformers can be viewed as a two-layer fast-weight MLP, whose weights are dynamically induced by inputs and whose hidden dimension is equal to the sequence length N . As the context extends, the expressive capacity of such an N -width MLP increases, but it becomes unscalable for extremely long sequences. Recently, this fast-weight perspective has motivated the Mixture-of-Experts (MoE) attention mechanism, which partitions the sequence into rigid blocks, treats them as fast-weight experts, and sparsely routes the tokens to them. In this paper, we elevate this perspective to a unifying framework for efficient attention mechanisms, interpreting them as making fast weights scalable through either routing or compression, and organizing them into a five-dimensional taxonomy. Then, we propose Mixture-of-Top-k Attention (MiTA), which employs a small set of landmark queries to gather top-k attended key-value pairs as queryaware and deformable routed experts, while compressing the N -width MLP into a narrower shared expert. Consequently, our MiTA improves the flexibility of prior MoE attention from rigid to deformable fast-weight experts, as well as the scalability of prior top-k attention from query-specific set to reusable top-k set. We conduct extensive experiments on vision tasks showing the superior effectiveness and efficiency of our MiTA, and also uncovering intriguing properties such as an emergent token-pruning effect and easy generalization from standard attention. Code is available at https://github.com/QishuaiWen/MiTA. Recent work along this direction has demonstrated that sparse attention inspired by Mixture-of-Experts (MoE) can scale up the effective sequence length, deliver wall-clock speedups, and accommodate large-scale pretraining [9,10]. In contrast to traditional MoE, whose experts are static and structured model parameters (i.e., slow weights), their experts are input-dependent fast weights [11], Preprint.

For agents

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

get_harvested_code_for_paper("2602.01219")
get_code_for_paper("2602.01219")
have("2602.01219")

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

curl https://syntology.ai/api/ran/2602.01219.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.01219.svg)](https://syntology.ai/paper/2602.01219)

Connect an agent — have() is free.