SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2412.06540 · NeurIPS · 2025

Sloth: scaling laws for LLM skills to predict multi-benchmark performance across families

Mikhail Yurochkin, Yuekai Sun, Leshem Choshen, Felipe Polo, Seamus Somerstep

arXiv · PDF · Open in the Atlas

Code that ran

8 of 9 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, 0 violated it and 0 have no contract verdict; 4 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/2412.06540.json · badge: /api/ran/2412.06540.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
felipemaiapolo/sloth canonical (classifier label)
classifier basis: census
8 ran · 1 no recorded run
FunctionStatusWhere it lives
filter Ran
ran; our contract draft was wrong, not the code
felipemaiapolo/sloth/sloth/utils.py
code served (permissive licence) · get_code("869a887d132f2470")
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 str. 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

def filter(s):
    try:
        s = s.split('/')[1]
    except:
        s = s
    try:
        s = s.split('__')[1]
    except:
        s = s
    return s.lower().replace('-hf', '').replace('_', '').replace('-', '')

ARGS = {
    's': 'model-name__some-identifier-hf'
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
forward1 Ran
ran, honoured the contract we drafted
felipemaiapolo/sloth/sloth/sloth.py
code served (permissive licence) · get_code("83a97028835dd404")
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 (2, 5) 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)

# Create small tensors for the matrix multiplication operation
# X @ W1_X requires X shape (batch, feat_x) and W1_X shape (feat_x, out)
# D @ W1_D requires D shape (batch, feat_d) and W1_D shape (feat_d, out)

batch_size = 2
feat_x = 4
feat_d = 3
output_dim = 5

X = torch.randn(batch_size, feat_x)
D = torch.randn(batch_size, feat_d)
W1_X = torch.randn(feat_x, output_dim)
W1_D = torch.randn(feat_d, output_dim)

ARGS = {
    "X": X,
    "D": D,
    "W1_X": W1_X,
    "W1_D": W1_D
}
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 (2, 5) — shape (2, 5) 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
forward2 Ran
ran, honoured the contract we drafted
felipemaiapolo/sloth/sloth/sloth.py
code served (permissive licence) · get_code("f975285e09aeea1e")
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 (2, 3) 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 numpy as np

torch.manual_seed(0)

# Z1: result of a previous layer, shape (batch, hidden_dim)
Z1 = torch.randn(2, 4)

# W2: weight matrix for second layer, shape (hidden_dim, output_dim)
W2 = torch.randn(4, 3)

# b2: bias vector for second layer, shape (output_dim,)
b2 = torch.randn(3)

ARGS = {
    "Z1": Z1,
    "W2": W2,
    "b2": b2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, 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 (2, 3) — shape (2, 3) 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_true_indices Ran
ran, honoured the contract we drafted
felipemaiapolo/sloth/sloth/utils.py
code served (permissive licence) · get_code("94abc955695450b9")
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 ndarray, ndim 1, dtype family int, shape (3,) 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 numpy as np

bool_array = np.array([True, False, True, False, True])

ARGS = {
    "bool_array": bool_array
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family int, shape (3,) — shape (3,) 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
logit Ran
ran; our contract draft was wrong, not the code
felipemaiapolo/sloth/sloth/sloth.py
code served (permissive licence) · get_code("c96121833ed7e023")
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 (2, 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)

# The logit function takes a tensor x and an optional eps parameter
# It computes log((x + eps) / (1 - x + eps))
# x should be a tensor with values in a valid range (typically (0, 1) for logit)

# Create a small tensor with values in the range (0, 1)
x = torch.tensor([[0.1, 0.5, 0.9, 0.3],
                   [0.2, 0.7, 0.4, 0.8]], dtype=torch.float32)

ARGS = {
    "x": x,
    "eps": 1e-05
}
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 (2, 4) — shape (2, 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
prep_data Ran
ran; our contract draft was wrong, not the code
felipemaiapolo/sloth/sloth/experiment.py
code served (permissive licence) · get_code("004484c84acb696d")
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 pandas as pd
import numpy as np

# Create a minimal DataFrame matching the function's expected columns
data = pd.DataFrame({
    'Pretraining Data Size (T)': [1e6, 2e6, 3e6, 4e6, 5e6, 6e6],
    '#Params (B)': [1.0, 2.0, 3.0, 1.5, 2.5, 3.5],
    'FLOPs (1E21)': [100.0, 200.0, 300.0, 150.0, 250.0, 350.0],
    'Family': ['ModelA', 'ModelA', 'ModelA', 'ModelB', 'ModelB', 'ModelB'],
    'Instruct': [0.5, 0.6, 0.7, 0.55, 0.65, 0.75],
    'Benchmark1': [10.0, 11.0, 12.0, 10.5, 11.5, 12.5],
    'Benchmark2': [20.0, 21.0, 22.0, 20.5, 21.5, 22.5],
})

benchs_names = ['Benchmark1', 'Benchmark2']
min_models = 2

ARGS = {
    'data': data,
    'benchs_names': benchs_names,
    'min_models': min_models,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas — 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
prep_data2 Ran
ran; our contract draft was wrong, not the code
felipemaiapolo/sloth/sloth/experiment.py
code served (permissive licence) · get_code("a9375f45a421c52a")
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 numpy as np
import pandas as pd
from sklearn.preprocessing import LabelEncoder

# Create a minimal DataFrame with required columns
np.random.seed(0)

# Define the families and benchmark names
families = ['fam_A', 'fam_B', 'fam_C']
benchs_names = ['bench1', 'bench2']
test_families = ['fam_B', 'fam_C']

# Create sample data with all required columns
n_rows = 12
data = pd.DataFrame({
    'family': np.repeat(families, n_rows // len(families)),
    'logT': np.random.randn(n_rows),
    'logS': np.random.randn(n_rows),
    'logSlogT': np.random.randn(n_rows),
    'logS2': np.random.randn(n_rows),
    'logT2': np.random.randn(n_rows),
    'logF': np.random.randn(n_rows),
    'bench1': np.random.randn(n_rows),
    'bench2': np.random.randn(n_rows),
    'Instruct': ['instr_' + str(i) for i in range(n_rows)],
})

ARGS = {
    'data': data,
    'test_families': test_families,
    'benchs_names': benchs_names,
    'n_train_models': 1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, sklearn — 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
sigmoid_np Ran
ran, honoured the contract we drafted
felipemaiapolo/sloth/sloth/utils.py
code served (permissive licence) · get_code("5a36846dab98ce35")
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 ndarray, ndim 2, dtype family float, shape (2, 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 numpy as np

def sigmoid_np(z):
    return 1 / (1 + np.exp(-z))

# Construct a small numpy array to pass to sigmoid_np
z = np.array([[-2.0, -1.0, 0.0, 1.0],
              [2.0, 0.5, -0.5, 1.5]])

ARGS = {
    "z": z
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
run_exp Not yet run
unverified (harvested, no recorded run)
felipemaiapolo/sloth/sloth/experiment.py
code served (permissive licence) · get_code("38473214bda25f09")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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 9 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

Scaling laws for large language models (LLMs) predict model performance based on parameters like size and training data. However, differences in training configurations and data processing across model families lead to significant variations in benchmark performance, making it difficult for a single scaling law to generalize across all LLMs. On the other hand, training family-specific scaling laws requires training models of varying sizes for every family. In this work, we propose Skills Scaling Laws (SSLaws, pronounced as Sloth), a novel scaling law that leverages publicly available benchmark data and assumes LLM performance is driven by low-dimensional latent skills, such as reasoning and instruction following. These latent skills are influenced by computational resources like model size and training tokens, but with varying efficiencies across model families. Sloth exploits correlations across benchmarks to provide more accurate and interpretable predictions while alleviating the need to train multiple LLMs per family. We present both theoretical results on parameter identification and empirical evaluations on 12 prominent benchmarks, from Open LLM Leaderboard v1/v2, demonstrating that Sloth predicts LLM performance accurately and offers insights into scaling behaviors for complex downstream tasks, increased test-time compute, and compute-optimal scaling of skills. Our code can be found on https://github.com/felipemaiapolo/sloth.

For agents

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

get_harvested_code_for_paper("2412.06540")
get_code_for_paper("2412.06540")
have("2412.06540")

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

curl https://syntology.ai/api/ran/2412.06540.json

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

[![Syntology run record](https://syntology.ai/api/ran/2412.06540.svg)](https://syntology.ai/paper/2412.06540)

Connect an agent — have() is free.