SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.08656 · NeurIPS · 2020

Multiscale Deep Equilibrium Models

J Zico Kolter, Vladlen Koltun, Shaojie Bai

arXiv · PDF · Open in the Atlas

Code that ran

10 of 14 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (6 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 7 have no contract verdict; 3 failures of our instrument; 4 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/2006.08656.json · badge: /api/ran/2006.08656.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
locuslab/mdeq canonical (classifier label)
classifier basis: classifier_2025
3 ran · 1 no recorded run
locuslab/deq community
Papers with Code archive
5 ran · 1 no recorded run
lufanma/ifr community
Papers with Code archive
1 ran · 2 no recorded run
SinclairHudson/DeepEquilibrium community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
BranchNet Ran
ran on a synthesized input · constructed an object; no result was computed
locuslab/mdeq/lib/models/mdeq_core.py
code served (permissive licence) · get_code("16eed3f65715e8e9")
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, BranchNet: 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

torch.manual_seed(0)

class DummyBlock(nn.Module):
    """A minimal block that accepts x and optional injection"""
    def __init__(self, in_dim, out_dim):
        super().__init__()
        self.linear = nn.Linear(in_dim, out_dim)
    
    def forward(self, x, injection=None):
        return self.linear(x)

class DummyBlockNoInjection(nn.Module):
    """A minimal block that only accepts x (for blocks after the first)"""
    def __init__(self, in_dim, out_dim):
        super().__init__()
        self.linear = nn.Linear(in_dim, out_dim)
    
    def forward(self, x):
        return self.linear(x)

# Create blocks list with proper dimensions
# First block accepts injection parameter
block1 = DummyBlock(in_dim=4, out_dim=5)
# Subsequent blocks only take x
block2 = DummyBlockNoInjection(in_dim=5, out_dim=6)
block3 = DummyBlockNoInjection(in_dim=6, out_dim=4)

blocks = [block1, block2, block3]

ARGS = {
    "blocks": blocks
}
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 BranchNet — 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 BranchNet 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
DEQ Ran
ran on a synthesized input · constructed an object; no result was computed
SinclairHudson/DeepEquilibrium/DEQ.py
pointer only (licence: NONE) · get_code("ecde83d001c2ad88")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, SinclairHudson/DeepEquilibrium when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

DownsampleModule Ran
ran on a synthesized input · constructed an object; no result was computed
locuslab/mdeq/lib/models/mdeq_core.py
code served (permissive licence) · get_code("68194e5af1ed3376")
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, DownsampleModule: 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 collections import OrderedDict

torch.manual_seed(0)

NUM_GROUPS = 4

# Define num_channels as a dict mapping resolutions to channel counts
# in_res and out_res are indices into this dict
num_channels = {
    0: 4,
    1: 8,
    2: 16,
    3: 32
}

# Define parameters
in_res = 0
out_res = 2

ARGS = {
    'num_channels': num_channels,
    'in_res': in_res,
    'out_res': out_res
}
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 DownsampleModule — 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 DownsampleModule 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
DownsampleModule Ran
ran on a synthesized input
locuslab/deq/MDEQ-Vision/lib/models/mdeq_core.py
code served (permissive licence) · get_code("94faf9823bec3580")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 4, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 collections import OrderedDict

FUSE_GN_AFFINE = True
NUM_GROUPS = 4

class DownsampleModule(nn.Module):

    def __init__(self, num_channels, in_res, out_res):
        """
        A downsample step from resolution j (with in_res) to resolution i (with out_res). A series of 2-strided convolutions.
        """
        super(DownsampleModule, self).__init__()
        convs = []
        inp_chan = num_channels[in_res]
        out_chan = num_channels[out_res]
        self.level_diff = level_diff = out_res - in_res
        kwargs = {'kernel_size': 3, 'stride': 2, 'padding': 1, 'bias': False}
        for k in range(level_diff):
            intermediate_out = out_chan if k == level_diff - 1 else inp_chan
            components = [('conv', nn.Conv2d(inp_chan, intermediate_out, **kwargs)), ('gnorm', nn.GroupNorm(NUM_GROUPS, intermediate_out, affine=FUSE_GN_AFFINE))]
            if k != level_diff - 1:
                components.append(('relu', nn.ReLU(inplace=True)))
            convs.append(nn.Sequential(OrderedDict(components)))
        self.net = nn.Sequential(*convs)

    def forward(self, x):
        return self.net(x)


INIT_ARGS = {
    'num_channels': {0: 4, 1: 8, 2: 16},
    'in_res': 0,
    'out_res': 2,
}

ARGS = {
    'x': torch.randn(2, 4, 32, 32),
}

if __name__ == '__main__':
    module = DownsampleModule(**INIT_ARGS)
    output = module(**ARGS)
    print(f"Output shape: {output.shape}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 4, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
UpsampleModule Ran
ran on a synthesized input · constructed an object; no result was computed
locuslab/mdeq/lib/models/mdeq_core.py
code served (permissive licence) · get_code("2b22e865b09bfe0c")
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, UpsampleModule: 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 collections import OrderedDict

torch.manual_seed(0)

# The UpsampleModule constructor takes:
# - num_channels: a dict mapping resolution indices to channel counts
# - in_res: input resolution index
# - out_res: output resolution index

# Looking at the body:
# - num_channels[in_res] and num_channels[out_res] are accessed to get channel counts
# - level_diff = in_res - out_res is computed
# - Conv2d expects (inp_chan, out_chan, kernel_size=1, bias=False)
# - GroupNorm expects (NUM_GROUPS, out_chan, affine=True)
# - Upsample expects scale_factor=2**level_diff

# Create a num_channels dict with distinct channel values for different resolutions
num_channels = {
    0: 16,
    1: 32,
    2: 64,
    3: 128,
    4: 256,
}

in_res = 2
out_res = 0

ARGS = {
    'num_channels': num_channels,
    'in_res': in_res,
    'out_res': out_res,
}
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 UpsampleModule — 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 UpsampleModule 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
UpsampleModule Ran
ran on a synthesized input · constructed an object; no result was computed
locuslab/deq/MDEQ-Vision/lib/models/mdeq_core.py
code served (permissive licence) · get_code("f945eaf4c694d1fe")
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, UpsampleModule: 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 collections import OrderedDict

torch.manual_seed(0)

# num_channels is a dict mapping resolution levels to channel counts
num_channels = {
    0: 8,   # in_res = 0
    1: 16,  # out_res = 1
    2: 32,
    3: 64
}

# in_res and out_res are resolution indices
# The forward method will receive a tensor and pass it through the network
# The network expects input channels matching num_channels[in_res]
# and outputs channels matching num_channels[out_res]

in_res = 2
out_res = 1

ARGS = {
    'num_channels': num_channels,
    'in_res': in_res,
    'out_res': out_res
}
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 UpsampleModule — 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 UpsampleModule 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
WeightNorm Ran
ran on a synthesized input · constructed an object; no result was computed
locuslab/deq/MDEQ-Vision/lib/models/mdeq_core.py
code served (permissive licence) · get_code("711a21f05476c8a1")
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, WeightNorm: 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 torch.nn.parameter import Parameter

torch.manual_seed(0)

# Create a simple module with a weight parameter
class SimpleModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.weight = Parameter(torch.randn(4, 3))
        self.bias = Parameter(torch.randn(3))
    
    def forward(self, x):
        return x @ self.weight + self.bias

module = SimpleModule()

# Define arguments for WeightNorm.__init__
names = ['weight']
dim = 0

ARGS = {
    'names': names,
    'dim': dim
}
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 WeightNorm — 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 WeightNorm 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
_norm Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
locuslab/deq/MDEQ-Vision/lib/models/mdeq_core.py
code served (permissive licence) · get_code("a09788a6b90f7c7c")
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 0, dtype family float, shape () 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)

# Test with a multi-dimensional tensor
# Create a tensor with shape (2, 3, 4, 5) to test different dim cases
p = torch.randn(2, 3, 4, 5)

# Test case 1: dim=None - computes norm over all dimensions
ARGS = {
    "p": p,
    "dim": None
}
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 0, dtype family float, shape () — shape () 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
analyze_broyden Ran
ran; our contract draft was wrong, not the code
lufanma/ifr/lib/modules/deq2d.py
pointer only (licence: NONE) · get_code("ed88fb917332b540")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, lufanma/ifr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

weight_norm Ran
ran; our contract draft was wrong, not the code
locuslab/deq/MDEQ-Vision/lib/models/mdeq_core.py
code served (permissive licence) · get_code("a27206c36724f1c4")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim 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 torch.nn.parameter import Parameter

torch.manual_seed(0)

# Create a simple module with a weight parameter
class SimpleModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.weight = nn.Parameter(torch.randn(4, 3))
        self.bias = nn.Parameter(torch.randn(3))
    
    def forward(self, x):
        return x @ self.weight + self.bias

# Instantiate the module
module = SimpleModule()

# Define arguments for weight_norm function
ARGS = {
    'module': module,
    'names': ['weight'],
    'dim': 0
}
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 tuple, ndim 1, dtype family float — a tuple output has no array shape of its own; none is stated
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
DEQFunc2d Not yet run
unverified (harvested, no recorded run)
lufanma/ifr/lib/modules/deq2d.py
pointer only (licence: NONE) · get_code("625bb3ec12707aaa")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, lufanma/ifr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

DEQModule2d Not yet run
unverified (harvested, no recorded run)
lufanma/ifr/lib/modules/deq2d.py
pointer only (licence: NONE) · get_code("bf6157640e178703")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, lufanma/ifr when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

MDEQModule Not yet run
unverified (harvested, no recorded run)
locuslab/mdeq/lib/models/mdeq_core.py
code served (permissive licence) · get_code("96aa0f0d7c9dc48a")
How it ran

no invocation recorded.

MDEQModule Not yet run
unverified (harvested, no recorded run)
locuslab/deq/MDEQ-Vision/lib/models/mdeq_core.py
code served (permissive licence) · get_code("701031360acd0728")
How it ran

no invocation recorded.

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 2 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 2 later) and 12 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
weight_norm
brick of this paper's assembly: helper
5 other papers
1 dated earlier, 4 later
1810.06682 (2018-10)
Trellis Networks for Sequence Modeling
2111.05177 (2021-11)
On Training Implicit Models
2308.10873 (2023-08)
SpikingBERT: Distilling BERT to Train Spiking Language Models Using Implicit Differentiation
and 2 more not named here
analyze_broyden
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2602.03297 (2026-02)
Lipschitz Multiscale Deep Equilibrium Models: A Theoretically Guaranteed and Accelerated Approach

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

We propose a new class of implicit networks, the multiscale deep equilibrium model (MDEQ), suited to large-scale and highly hierarchical pattern recognition domains. An MDEQ directly solves for and backpropagates through the equilibrium points of multiple feature resolutions simultaneously, using implicit differentiation to avoid storing intermediate states (and thus requiring only O(1) memory consumption). These simultaneously-learned multi-resolution features allow us to train a single model on a diverse set of tasks and loss functions, such as using a single MDEQ to perform both image classification and semantic segmentation. We illustrate the effectiveness of this approach on two large-scale vision tasks: ImageNet classification and semantic segmentation on high-resolution images from the Cityscapes dataset. In both settings, MDEQs are able to match or exceed the performance of recent competitive computer vision models: the first time such performance and scale have been achieved by an implicit deep learning approach. The code and pre-trained models are at https://github.com/locuslab/mdeq.

For agents

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

get_harvested_code_for_paper("2006.08656")
get_code_for_paper("2006.08656")
have("2006.08656")

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

curl https://syntology.ai/api/ran/2006.08656.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.08656.svg)](https://syntology.ai/paper/2006.08656)

Connect an agent — have() is free.