SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.22113 · NeurIPS · 2024

Where Do Large Learning Rates Lead Us?

Ekaterina Lobacheva, Dmitry Vetrov, Maxim Kodryan, Ildus Sadrtdinov, Eduard Pokonechny

arXiv · PDF · Open in the Atlas

Code that ran

11 of 16 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; 4 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 6 failures of our instrument; 5 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/2410.22113.json · badge: /api/ran/2410.22113.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
isadrtdinov/understanding-large-lrs canonical (classifier label)
classifier basis: census
11 ran · 5 no recorded run
FunctionStatusWhere it lives
LogSumExp Ran
ran; our contract draft was wrong, not the code
isadrtdinov/understanding-large-lrs/convolutional_networks/training_utils.py
code served (permissive licence) · get_code("8cb1d24a1cf29e11")
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 (1, 4) 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)

# Create a small tensor for testing LogSumExp
# The function expects a tensor and a dim parameter
x = torch.randn(2, 4)  # batch_size=2, feature_dim=4

ARGS = {
    "x": x,
    "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 Tensor, ndim 2, dtype family float, shape (1, 4) — shape (1, 4) 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
Lop Ran
ran; our contract draft was wrong, not the code
isadrtdinov/understanding-large-lrs/convolutional_networks/get_info_funcs.py
code served (permissive licence) · get_code("19b54a137a0bbdb7")
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 3, dtype family float; no output shape is stated. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:58:14Z. 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

torch.manual_seed(0)

# Create input tensors
# ys: output of a computation that requires grad
xs = torch.tensor([[1.0, 2.0, 3.0, 4.0],
                    [5.0, 6.0, 7.0, 8.0]], requires_grad=True)

# Create a simple computation graph
ys = (xs ** 2).sum()

# ws: gradient outputs weights, should match shape of ys
# Since ys is a scalar, ws should be a scalar
ws = torch.tensor(1.0)

ARGS = {
    "ys": ys,
    "xs": xs,
    "ws": ws
}
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 3, 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 syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); 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
check_si_name Ran
ran, violated the contract we drafted
isadrtdinov/understanding-large-lrs/convolutional_networks/get_info.py
code served (permissive licence) · get_code("d2c03e2aaa143601")
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 bool, 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)

ARGS = {
    'n': 'conv1',
    'model_name': 'ResNet18'
}
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 bool, 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
cross_entropy Ran
ran; our contract draft was wrong, not the code
isadrtdinov/understanding-large-lrs/convolutional_networks/custom_training_utils.py
code served (permissive licence) · get_code("92278c6d1c2ed4ba")
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. 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)

# Build a minimal model that takes input and produces class logits
class SimpleClassifier(nn.Module):
    def __init__(self, input_dim, num_classes):
        super().__init__()
        self.linear = nn.Linear(input_dim, num_classes)
    
    def forward(self, x):
        return self.linear(x)

# Create model with 4 input features and 3 output classes
model = SimpleClassifier(input_dim=4, num_classes=3)

# Create tiny batch: batch_size=2, input_dim=4
input_tensor = torch.randn(2, 4)

# Create target labels for 2 samples with 3 classes
target_tensor = torch.tensor([0, 2])

ARGS = {
    'model': model,
    'input': input_tensor,
    'target': target_tensor,
    'reduction': 'mean'
}
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 — a tuple output has no array shape of its own; none is 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
flatten Ran
ran, honoured the contract we drafted
isadrtdinov/understanding-large-lrs/convolutional_networks/training_utils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from ubc-tea/local-superior-soups, is pointer only (licence: NONE) · get_code("cdc64e98aaf2dd54")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from ubc-tea/local-superior-soups, and GitHub detected no licence in that repository 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.

get_grads Ran
ran, honoured the contract we drafted
isadrtdinov/understanding-large-lrs/synthetic_example/model.py
code served (permissive licence) · get_code("84a8ef25008d233d")
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 1, dtype family float, shape (23,) 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
import torch.nn as nn

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

# Create model and dummy data
model = SimpleModel()
x = torch.randn(2, 4)
y = torch.randn(2, 2)

# Perform a forward pass and compute gradients
output = model(x)
loss = torch.nn.functional.mse_loss(output, y)
loss.backward()

ARGS = {
    "model": model
}
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 1, dtype family float, shape (23,) — shape (23,) 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
get_weights Ran
ran, honoured the contract we drafted
isadrtdinov/understanding-large-lrs/synthetic_example/model.py
code served (permissive licence) · get_code("79a3e75a15ed94dd")
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 1, dtype family float, shape (23,) 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
import torch.nn as nn

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

model = SimpleModel()

ARGS = {
    "model": model
}
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 1, dtype family float, shape (23,) — shape (23,) 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
interpolate_state_dicts Ran
ran; our contract draft was wrong, not the code
isadrtdinov/understanding-large-lrs/convolutional_networks/custom_training_utils.py
code served (permissive licence) · get_code("6d9487d53006a554")
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 OrderedDict. 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 copy import deepcopy

torch.manual_seed(0)

# Create two simple model state dicts
# These should have the same structure since we're interpolating between them

model_A = nn.Sequential(
    nn.Linear(4, 3),
    nn.ReLU(),
    nn.Linear(3, 2)
)

model_B = nn.Sequential(
    nn.Linear(4, 3),
    nn.ReLU(),
    nn.Linear(3, 2)
)

# Get state dicts
model_A_state_dict = model_A.state_dict()
model_B_state_dict = model_B.state_dict()

# Create tensors with different values in model_B to make interpolation meaningful
for key in model_B_state_dict.keys():
    model_B_state_dict[key] = model_B_state_dict[key] + 1.0

ARGS = {
    "model_A_state_dict": model_A_state_dict,
    "model_B_state_dict": model_B_state_dict,
    "t": 0.5
}
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 OrderedDict — a OrderedDict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a OrderedDict 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
load_state_dict_from_swa_to_usual Ran
ran; our contract draft was wrong, not the code
isadrtdinov/understanding-large-lrs/convolutional_networks/custom_training_utils.py
code served (permissive licence) · get_code("01063d5c587678c2")
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 OrderedDict. 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
from collections import OrderedDict
import tempfile
import os

torch.manual_seed(0)

# Create a temporary file with a valid checkpoint
temp_dir = tempfile.mkdtemp()
temp_path = os.path.join(temp_dir, "checkpoint.pt")

# Create a state dict with module-prefixed keys and n_averaged
state_dict = OrderedDict([
    ('module.layer1.weight', torch.randn(4, 3)),
    ('module.layer1.bias', torch.randn(4)),
    ('module.layer2.weight', torch.randn(5, 4)),
    ('n_averaged', torch.tensor(10)),
])

# Save as a checkpoint with 'state_dict' key
checkpoint = {'state_dict': state_dict}
torch.save(checkpoint, temp_path)

ARGS = {
    'path': temp_path
}
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 OrderedDict — a OrderedDict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a OrderedDict 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
num_params Ran
ran, honoured the contract we drafted
isadrtdinov/understanding-large-lrs/synthetic_example/model.py
code served (permissive licence) · get_code("d28ce1ac93c376f2")
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 int, ndim 0, dtype family int; no output shape is stated. It ran in syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1) at 2026-09-27T00:58:14Z. 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

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        return x

model = SimpleModel()

ARGS = {
    "model": model
}
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 int, ndim 0, dtype family int — no output shape is stated: no later run recorded an output shape for this sample (observed_output and metamorphic_properties carry none; shape_contract and output_contract record ndim only)
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 syntology-harvest-sandbox:v5-deps (sha256:6a292aa1aed1); 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
unflatten_like Ran
ran; our contract draft was wrong, not the code
isadrtdinov/understanding-large-lrs/convolutional_networks/get_info_funcs.py
code served (permissive licence) · get_code("6d8daee3396e21d5")
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 list. 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)

# Create a flattened vector
vector = torch.randn(15)

# Create a list of tensors with different shapes to unflatten like
likeTensorList = [
    torch.randn(2, 3),      # 6 elements
    torch.randn(4, 2),      # 8 elements
    torch.randn(1),         # 1 element
]

ARGS = {
    "vector": vector,
    "likeTensorList": likeTensorList,
}
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 list — a list output has no array shape of its own; none is 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
Rop Not yet run
unverified (harvested, no recorded run)
isadrtdinov/understanding-large-lrs/convolutional_networks/get_info_funcs.py
code served (permissive licence) · get_code("8bc49ef2a0810046")
How it ran

no invocation recorded.

get_model_extra Not yet run
unverified (harvested, no recorded run)
isadrtdinov/understanding-large-lrs/convolutional_networks/nb_utils.py
code served (permissive licence) · get_code("157cf0b88320c24e")
How it ran

no invocation recorded.

get_model_str Not yet run
unverified (harvested, no recorded run)
isadrtdinov/understanding-large-lrs/convolutional_networks/nb_utils.py
code served (permissive licence) · get_code("c415f0e6392ad2af")
How it ran

no invocation recorded.

svhn_loaders Not yet run
unverified (harvested, no recorded run)
isadrtdinov/understanding-large-lrs/convolutional_networks/data.py
code served (permissive licence) · get_code("182de3ac9284f86e")
How it ran

no invocation recorded.

unflatten_like Not yet run
unverified (harvested, no recorded run)
isadrtdinov/understanding-large-lrs/convolutional_networks/training_utils.py
code served (permissive licence) · get_code("2da33912415648f3")
How it ran

no invocation recorded.

Also present in other papers' code

Of 16 samples lifted from this paper's linked repositories, 16 carry a function identity and 0 do not. Of the 16 distinct functions with an identity, 10 are also present in other papers whose code we hold (10 with at least one paper dated earlier, 1 the same month, 1 later) and 6 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
flatten 5 other papers
3 dated earlier, 1 the same month, 1 later
1703.04730 (2017-03)
Understanding Black-box Predictions via Influence Functions
2312.16979 (2023-12)
BlackboxBench: A Comprehensive Benchmark of Black-box Adversarial Attacks
2405.03425 (2024-05)
Gaussian Stochastic Weight Averaging for Bayesian Low-Rank Adaptation of Large Language Models
and 2 more not named here
unflatten_like 3 other papers
3 dated earlier
2106.15739 (2021-06)
On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
2209.03695 (2022-09)
Training Scale-Invariant Neural Networks on the Sphere Can Happen in Three Regimes
2312.17173 (2023-12)
Non-Vacuous Generalization Bounds for Large Language Models
Lop 2 other papers
2 dated earlier
2106.15739 (2021-06)
On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
2209.03695 (2022-09)
Training Scale-Invariant Neural Networks on the Sphere Can Happen in Three Regimes
Rop 2 other papers
2 dated earlier
2106.15739 (2021-06)
On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
2209.03695 (2022-09)
Training Scale-Invariant Neural Networks on the Sphere Can Happen in Three Regimes
check_si_name 2 other papers
2 dated earlier
2106.15739 (2021-06)
On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
2209.03695 (2022-09)
Training Scale-Invariant Neural Networks on the Sphere Can Happen in Three Regimes
get_model_extra 2 other papers
2 dated earlier
2106.15739 (2021-06)
On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
2209.03695 (2022-09)
Training Scale-Invariant Neural Networks on the Sphere Can Happen in Three Regimes
get_model_str 2 other papers
2 dated earlier
2106.15739 (2021-06)
On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
2209.03695 (2022-09)
Training Scale-Invariant Neural Networks on the Sphere Can Happen in Three Regimes
svhn_loaders 2 other papers
2 dated earlier
2106.15739 (2021-06)
On the Periodic Behavior of Neural Network Training with Batch Normalization and Weight Decay
2209.03695 (2022-09)
Training Scale-Invariant Neural Networks on the Sphere Can Happen in Three Regimes
unflatten_like 2 other papers
2 dated earlier
1703.04730 (2017-03)
Understanding Black-box Predictions via Influence Functions
2405.03425 (2024-05)
Gaussian Stochastic Weight Averaging for Bayesian Low-Rank Adaptation of Large Language Models
LogSumExp 1 other paper
1 dated earlier
1703.04730 (2017-03)
Understanding Black-box Predictions via Influence Functions

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

It is generally accepted that starting neural networks training with large learning rates (LRs) improves generalization. Following a line of research devoted to understanding this effect, we conduct an empirical study in a controlled setting focusing on two questions: 1) how large an initial LR is required for obtaining optimal quality, and 2) what are the key differences between models trained with different LRs? We discover that only a narrow range of initial LRs slightly above the convergence threshold lead to optimal results after fine-tuning with a small LR or weight averaging. By studying the local geometry of reached minima, we observe that using LRs from this optimal range allows for the optimization to locate a basin that only contains high-quality minima. Additionally, we show that these initial LRs result in a sparse set of learned features, with a clear focus on those most relevant for the task. In contrast, starting training with too small LRs leads to unstable minima and attempts to learn all features simultaneously, resulting in poor generalization. Conversely, using initial LRs that are too large fails to detect a basin with good solutions and extract meaningful patterns from the data. 1. Large LRs are preferred but how large are we talking about? 2. What are the key characteristics of the models trained with different LRs? * Equal contribution. † Shared senior authorship. 38th Conference on Neural Information Processing Systems (NeurIPS 2024).

For agents

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

get_harvested_code_for_paper("2410.22113")
get_code_for_paper("2410.22113")
have("2410.22113")

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

curl https://syntology.ai/api/ran/2410.22113.json

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

[![Syntology run record](https://syntology.ai/api/ran/2410.22113.svg)](https://syntology.ai/paper/2410.22113)

Connect an agent — have() is free.