SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.12031 · NeurIPS · 2024

Large Scale Transfer Learning for Tabular Data via Language Modeling

Juan Perdomo, Ludwig Schmidt, Josh Gardner

arXiv · PDF · Open in the Atlas

Code that ran

13 of 15 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; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 12 failures of our instrument; 2 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/2406.12031.json · badge: /api/ran/2406.12031.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
mlfoundations/tabliblib canonical (classifier label)
classifier basis: census
8 ran · 0 no recorded run
mlfoundations/rtfm canonical (classifier label)
classifier basis: census
5 ran · 2 no recorded run
FunctionStatusWhere it lives
add_content_hash Ran
ran; our contract draft was wrong, not the code
mlfoundations/tabliblib/tabliblib/mappers.py
code served (permissive licence) · get_code("4beae4af6e4059ec")
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 dict. 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 hashlib
import sys
from typing import Dict, Any
import numpy as np

# Construct a minimal valid row dict that satisfies the function's requirements
row = {
    'key': 'path/to/some/hash_value_abc123xyz',
}

ARGS = {
    'row': row,
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
arrow_to_pandas_safe Ran
ran; our contract draft was wrong, not the code
mlfoundations/tabliblib/tabliblib/io.py
code served (permissive licence) · get_code("5af725e76bd12300")
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 DataFrame. 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 pyarrow as pa
import pandas as pd
import numpy as np

# Create a simple valid PyArrow table
data = {
    'col1': [1, 2, 3],
    'col2': ['a', 'b', 'c'],
    'col3': [1.1, 2.2, 3.3]
}
arrow_table = pa.table(data)

ARGS = {
    't': arrow_table,
    'raise_on_error': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, pyarrow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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
batch_to_xpu Ran
ran; our contract draft was wrong, not the code
mlfoundations/rtfm/rtfm/torch_utils.py
code served (permissive licence) · get_code("1af054b95295710e")
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 dict. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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

torch.manual_seed(0)

batch = {
    "input_ids": torch.tensor([[1, 2, 3, 4], [5, 6, 7, 8]], dtype=torch.long),
    "attention_mask": torch.tensor([[1, 1, 1, 0], [1, 1, 1, 1]], dtype=torch.long),
    "token_type_ids": torch.tensor([[0, 0, 1, 1], [0, 1, 1, 1]], dtype=torch.long),
}

ARGS = {
    "batch": batch,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
accelerate, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
drop_duplicates Ran
ran; our contract draft was wrong, not the code
mlfoundations/tabliblib/tabliblib/mappers.py
code served (permissive licence) · get_code("0ad24980bda6f6e4")
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 dict. 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

# Create a simple batch dictionary with multiple entries, some with duplicate content_hash
batch = {
    'content_hash': np.array(['hash1', 'hash2', 'hash1', 'hash3']),
    'path': np.array(['path/a', 'path/b', 'path/c', 'path/d']),
    'data': np.array([1, 2, 3, 4]),
    'sortkey1': np.array([10, 20, 30, 40], dtype=np.uint64),
    'sortkey2': np.array([100, 200, 300, 400], dtype=np.uint64),
}

ARGS = {
    'batch': batch,
    'key': 'content_hash',
    'sort_by': 'path',
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
example_ids_to_attention_mask Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mlfoundations/rtfm/rtfm/data.py
code served (permissive licence) · get_code("74807ffb0cddac16")
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 (9, 9) 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
from typing import List

torch.manual_seed(0)

# The function expects a list of integers representing example_ids
# The example_ids should contain values from 0 to max_example_id
# For a simple test case, let's create a sequence with 3 different example IDs
# with varying lengths to create blocks of different sizes

example_ids = [0, 0, 0, 1, 1, 2, 2, 2, 2]

ARGS = {
    "example_ids": example_ids
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — 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 (9, 9) — shape (9, 9) 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
find_covering_set Ran
ran; our contract draft was wrong, not the code
mlfoundations/tabliblib/tabliblib/dedup_utils.py
code served (permissive licence) · get_code("4767aa5be418958c")
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, ndim 2, 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.

from collections import defaultdict

# Analyze the function:
# - find_covering_set takes file_string_tuples parameter
# - It iterates over tuples of (file, string)
# - It builds defaultdicts and sets from these tuples
# - It performs a greedy set cover algorithm
# - No special methods or attributes are called on file or string objects
#   beyond standard operations (add, &, -=, extend, del)
# - file and string can be any hashable types (str, int, etc.)

# Create a small valid test case for the set cover problem
file_string_tuples = [
    ("file_a", "string_1"),
    ("file_a", "string_2"),
    ("file_b", "string_2"),
    ("file_b", "string_3"),
    ("file_c", "string_1"),
    ("file_c", "string_3"),
]

ARGS = {
    "file_string_tuples": file_string_tuples,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 2, dtype family float — a list 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
generate_random_dates Ran
ran; our contract draft was wrong, not the code
mlfoundations/tabliblib/tabliblib/summarizers.py
code served (permissive licence) · get_code("8d749d50f25dd3c2")
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, 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 numpy as np
from datetime import datetime, timedelta

ARGS = {
    'start_date': datetime(2000, 1, 1),
    'end_date': datetime(2023, 12, 31)
}
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 list, ndim 1, dtype family float — a list 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
get_categorical_idxs Ran
ran, honoured the contract we drafted
mlfoundations/rtfm/rtfm/tree_baselines.py
code served (permissive licence) · get_code("cd4c51c2326006e9")
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, ndim 1, dtype family int; 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 numpy as np
import pandas as pd

# Create a small DataFrame with mixed dtypes
# Some numeric columns, some categorical (object) columns, and a bool column
df = pd.DataFrame({
    'numeric_int': [1, 2],
    'numeric_float': [1.5, 2.5],
    'categorical_str': ['a', 'b'],
    'categorical_obj': ['x', 'y'],
    'bool_col': [True, False]
})

ARGS = {
    'df': df
}
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 list, ndim 1, dtype family int — a list 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
get_latest_checkpoint Ran
ran; our contract draft was wrong, not the code
mlfoundations/rtfm/rtfm/utils.py
code served (permissive licence) · get_code("ba26941c52b7f85a")
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 tempfile
import os
import time

# Create a temporary directory with checkpoint subdirectories
tmpdir = tempfile.mkdtemp()

# Create multiple checkpoint directories with different modification times
step_dirs = []
for i in range(3):
    step_dir = os.path.join(tmpdir, f'step_{i:04d}')
    os.makedirs(step_dir, exist_ok=True)
    step_dirs.append(step_dir)
    time.sleep(0.01)  # Ensure different modification times

ARGS = {
    'dirpath': tmpdir
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — 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
load_model_from_checkpoint Ran
ran; our contract draft was wrong, not the code
mlfoundations/rtfm/rtfm/train_utils.py
code served (permissive licence) · get_code("d718f67e89480b5b")
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
import os
import tempfile

torch.manual_seed(0)

# Create a minimal model that has load_state_dict method
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create a temporary directory with checkpoint
temp_dir = tempfile.mkdtemp(prefix='step-42_')

# Create a minimal state dict and save it
model = SimpleModel()
state_dict = model.state_dict()
model_path = os.path.join(temp_dir, 'model.pt')
torch.save(state_dict, model_path)

# Create a fresh model instance for the function to load into
model_for_loading = SimpleModel()

ARGS = {
    'model': model_for_loading,
    'ckpt_dir': temp_dir,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
safetensors, torch, tqdm — 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
path_to_str Ran
ran; our contract draft was wrong, not the code
mlfoundations/tabliblib/tabliblib/dedup_utils.py
code served (permissive licence) · get_code("39954605fabc6757")
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 path_to_str(path: str) -> str:
    """Convert a path to a string without slashes."""
    return path.replace('/', '__').replace('.parquet', '')

ARGS = {
    'path': 'data/folder/file.parquet'
}
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
sample_columns_if_needed Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
mlfoundations/tabliblib/tabliblib/io.py
code served (permissive licence) · get_code("02efcf6df251ab61")
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 DataFrame. 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

np.random.seed(0)

# Create a test dataframe with more columns than max_cols
# This will test the sampling path
data = {f'col_{i}': np.random.randn(10) for i in range(600)}
df_large = pd.DataFrame(data)

# Create a test dataframe with fewer columns than max_cols
# This will test the pass-through path
data_small = {f'col_{i}': np.random.randn(10) for i in range(100)}
df_small = pd.DataFrame(data_small)

ARGS = {
    "df": df_large,
    "max_cols": 512
}
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 DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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
write_arrow_bytes Ran
ran; our contract draft was wrong, not the code
mlfoundations/tabliblib/tabliblib/io.py
code served (permissive licence) · get_code("61bc6ff6d93f37c7")
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 bytes. 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 io
import pandas as pd
import pyarrow as pa
import numpy as np

# Create a small sample DataFrame with various data types
df = pd.DataFrame({
    'int_col': [1, 2, 3],
    'float_col': [1.5, 2.5, 3.5],
    'str_col': ['a', 'b', 'c'],
    'bool_col': [True, False, True]
})

ARGS = {
    'df': df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, pyarrow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bytes — a bytes output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a bytes 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
bool_cols_to_categorical Not yet run
unverified (harvested, no recorded run)
mlfoundations/rtfm/rtfm/tree_baselines.py
code served (permissive licence) · get_code("5326e55483a6717d")
How it ran

no invocation recorded.

prepare_4d_attention_mask Not yet run
unverified (harvested, no recorded run)
mlfoundations/rtfm/rtfm/data.py
code served (permissive licence) · get_code("7d1cac2a511468ae")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 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 15 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

Tabular data -structured, heterogeneous, spreadsheet-style data with rows and columns -is widely used in practice across many domains. However, while recent foundation models have reduced the need for developing task-specific datasets and predictors in domains such as language modeling and computer vision, this transfer learning paradigm has not had similar impact in the tabular domain. In this work, we seek to narrow this gap and present TABULA-8B, a language model for tabular prediction. We define a process for extracting a large, high-quality training dataset from the TabLib corpus, proposing methods for tabular data filtering and quality control. Using the resulting dataset, which comprises over 2.1B rows from over 4M unique tables, we fine-tune a Llama 3-8B large language model (LLM) for tabular data prediction (classification and binned regression) using a novel packing and attention scheme for tabular prediction. Through evaluation across a test suite of 329 datasets, we find that TABULA-8B has zero-shot accuracy on unseen tables that is over 15 percentage points (pp) higher than random guessing, a feat that is not possible with existing state-of-the-art tabular prediction models (e.g. XGBoost, TabPFN). In the few-shot setting (1-32 shots), without any fine-tuning on the target datasets, TABULA-8B is 5-15 pp more accurate than XGBoost and TabPFN models that are explicitly trained on equal, or even up to 16× more data. We release our model, code, and data along with the publication of this paper. 1

For agents

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

get_harvested_code_for_paper("2406.12031")
get_code_for_paper("2406.12031")
have("2406.12031")

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

curl https://syntology.ai/api/ran/2406.12031.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.12031.svg)](https://syntology.ai/paper/2406.12031)

Connect an agent — have() is free.