SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.01560 · 2024

OpenMathInstruct-2: Accelerating AI for Math with Massive Open-Source Instruction Data

arXiv · PDF · Open in the Atlas

Code that ran

12 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, 1 violated it and 0 have no contract verdict; 10 failures of our instrument; 3 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/2410.01560.json · badge: /api/ran/2410.01560.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
NVIDIA/NeMo-Skills community (classifier label)
classifier basis not recorded on the link
12 ran · 3 no recorded run
FunctionStatusWhere it lives
calculate_chunk_indices Ran
ran, honoured the contract we drafted
NVIDIA/NeMo-Skills/nemo_skills/file_utils.py
code served (permissive licence) · get_code("2e7aafbc230cb972")
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 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 torch
import numpy as np

torch.manual_seed(0)

# This function takes three integer parameters that describe how to chunk a dataset
# num_samples: total number of samples
# num_chunks: number of chunks to divide into
# chunk_id: which chunk (0-indexed)

ARGS = {
    "num_samples": 100,
    "num_chunks": 4,
    "chunk_id": 0,
}
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 tuple, ndim 1, dtype family int — 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
determine_proof_status Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/code_execution/proof_utils.py
code served (permissive licence) · get_code("dfb61cd2ac2b564a")
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
import re
from typing import Any, Dict

# Construct a valid compiler_output dictionary
# The function expects: process_status, stdout, stderr keys

# Test case 1: Completed successfully without sorry
compiler_output_success = {
    'process_status': 'completed',
    'stdout': 'Compilation succeeded\nProof verified',
    'stderr': ''
}

# Test case 2: Timeout
compiler_output_timeout = {
    'process_status': 'timeout',
    'stdout': 'Still compiling...',
    'stderr': 'Time limit exceeded'
}

# Test case 3: Has sorry (incomplete proof)
compiler_output_sorry = {
    'process_status': 'completed',
    'stdout': 'Proof complete except for sorry',
    'stderr': 'Warning: sorry used in proof'
}

# Test case 4: Other process status
compiler_output_error = {
    'process_status': 'error',
    'stdout': 'Compilation failed',
    'stderr': 'Syntax error on line 5'
}

# Use the first test case as the main argument
ARGS = {
    'compiler_output': compiler_output_success
}
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
extract_code_output Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/code_execution/utils.py
code served (permissive licence) · get_code("b4bbfbf06bb92de6")
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 re

def setup():
    generation = """
    Some text here
    <output>
    This is the extracted content
    </output>
    More text after
    """
    
    code_output_begin = "<output>"
    code_output_end = "</output>"
    extract_all = False
    
    return {
        "generation": generation,
        "code_output_begin": code_output_begin,
        "code_output_end": code_output_end,
        "extract_all": extract_all
    }

ARGS = setup()
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
extract_code_to_execute Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/code_execution/utils.py
code served (permissive licence) · get_code("b75a1053974f21ce")
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 re
from typing import Tuple

def _extract_between_separators(generation: str, separators: Tuple[str, str], extract_all: bool=False):
    """Extracting all text between last occurrence of separators[0] and [1].

    If extract_all is True, returning a list with all occurrences of text between separators.
    """
    if extract_all:
        separators = [re.escape(sp) for sp in separators]
        pattern = f'{separators[0]}(.*?){separators[1]}'
        return re.findall(pattern, generation, re.DOTALL)
    return generation.split(separators[0])[-1].split(separators[1])[0]

def extract_code_to_execute(generation: str, code_begin: str, code_end: str, extract_all: bool=False):
    return _extract_between_separators(generation, [code_begin, code_end], extract_all)


# Test case 1: Single extraction
generation_single = "Some text <CODE>print('hello')</CODE> more text"
code_begin = "<CODE>"
code_end = "</CODE>"

# Test case 2: Multiple extractions
generation_multiple = "First <CODE>x = 1</CODE> middle <CODE>y = 2</CODE> end"

ARGS = {
    "generation": generation_single,
    "code_begin": code_begin,
    "code_end": code_end,
    "extract_all": False
}
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
extract_proof_only Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/code_execution/proof_utils.py
code served (permissive licence) · get_code("df5a88c92c8f57ce")
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 re

def extract_proof_only(lean_code: str) -> str:
    """Extract only the proof part from a Lean theorem/example.

    This function removes the theorem/example header and returns just the proof body.

    Args:
        lean_code: The full Lean code including theorem statement

    Returns:
        The proof part only (everything after ':=')
    """
    lines = lean_code.strip().splitlines()
    if not lines:
        return ''
    header_start_pattern = re.compile('^\\s*(theorem|example)\\b')
    header_start_idx = None
    for i, line in enumerate(lines):
        if header_start_pattern.match(line):
            header_start_idx = i
            break
    if header_start_idx is None:
        return lean_code.strip()
    header_end_idx = None
    for i in range(header_start_idx, len(lines)):
        if ':=' in lines[i]:
            header_end_idx = i
            break
    if header_end_idx is None:
        return lean_code.strip()
    header_line, after = lines[header_end_idx].split(':=', 1)
    proof_first_line = after.strip()
    if proof_first_line:
        proof_lines = [proof_first_line] + lines[header_end_idx + 1:]
    else:
        proof_lines = lines[header_end_idx + 1:]
    if proof_lines:
        first = proof_lines[0].lstrip()
        if first == 'by':
            proof_lines = proof_lines[1:]
        elif first.startswith('by '):
            proof_lines[0] = first[3:]
    return '\n'.join(proof_lines).rstrip()


# Setup code to construct valid arguments
lean_code_example_1 = """theorem add_comm (a b : ℕ) : a + b = b + a := by
  induction a with
  | zero => simp
  | succ n ih => rw [Nat.succ_add, Nat.add_succ, ih]"""

lean_code_example_2 = """example (x : ℕ) : x = x :=
by rfl"""

lean_code_example_3 = """theorem test : True := by
  trivial"""

ARGS = {
    "lean_code": lean_code_example_1
}
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
format_code_output Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/code_execution/utils.py
code served (permissive licence) · get_code("b2186409cb2b7dfb")
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

ARGS = {
    'execution_dict': {
        'process_status': 'Success',
        'stdout': 'Output line 1\nOutput line 2',
        'stderr': ''
    },
    'code_output_begin': '<execution>',
    'code_output_end': '</execution>',
    'code_output_format': 'llama',
    'remaining_code_executions': 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 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
get_logger_name Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/utils.py
code served (permissive licence) · get_code("364d75dfb259e939")
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
import numpy as np
from pathlib import Path

# This function takes a file path string and returns a logger name
# It checks if '/nemo_skills/' is in the path and formats accordingly

ARGS = {
    'file': '/home/user/nemo_skills/training/model.py'
}
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 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
import_score_module Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/evaluation/compute_group_score.py
code served (permissive licence) · get_code("28ecc1878ccced7c")
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 module. 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 importlib.util
import sys
import tempfile
import os

# Create a temporary Python module file for testing
temp_dir = tempfile.mkdtemp()
temp_module_path = os.path.join(temp_dir, 'test_score_module.py')

# Write a simple valid Python module to the temporary file
with open(temp_module_path, 'w') as f:
    f.write("""
# Test score module
def score_function(x):
    return x * 2

class ScoreModel:
    def __init__(self):
        self.value = 42
""")

# Argument for the function - use the file path we just created
ARGS = {
    'score_module': temp_module_path
}
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 module — a module output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a module 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
init_wandb Ran
ran, violated the contract we drafted
NVIDIA/NeMo-Skills/nemo_skills/utils.py
code served (permissive licence) · get_code("a2afdfc356abe1d1")
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 os
import tempfile

# Create a temporary directory for experiment logs
temp_dir = tempfile.mkdtemp()

ARGS = {
    'project': 'test_project',
    'name': 'test_run',
    'exp_dir': temp_dir,
    'verbose': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
wandb — 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
jload Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/file_utils.py
code served (permissive licence) · get_code("8c48543062658550")
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 io
import json
import tempfile
import os

# Create temporary JSONL files for testing
temp_dir = tempfile.mkdtemp()

# Create first test file with valid JSON lines
file1_path = os.path.join(temp_dir, 'test1.jsonl')
with open(file1_path, 'w', encoding='utf-8') as f:
    f.write('{"name": "Alice", "age": 30}\n')
    f.write('{"name": "Bob", "age": 25}\n')
    f.write('invalid json line\n')  # This will be skipped
    f.write('{"name": "Charlie", "age": 35}\n')

# Create second test file with valid JSON lines
file2_path = os.path.join(temp_dir, 'test2.jsonl')
with open(file2_path, 'w', encoding='utf-8') as f:
    f.write('{"name": "David", "age": 40}\n')
    f.write('{"name": "Eve", "age": 28}\n')

# Create a StringIO object for testing with IOBase
io_object = io.StringIO('{"name": "Frank", "age": 32}\n{"name": "Grace", "age": 29}\n')

ARGS = {
    'filepath': file1_path,
    'mode': 'r',
    'verbose': False,
}
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 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
load_metric_files Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/evaluation/compute_group_score.py
code served (permissive licence) · get_code("6f55869f2815d973")
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 json
import tempfile
import os

# Create temporary metric files
temp_dir = tempfile.mkdtemp()

# Create first metric file
metric_file_1 = os.path.join(temp_dir, "metrics_1.json")
with open(metric_file_1, 'w') as f:
    json.dump({"accuracy": 0.95, "loss": 0.05}, f)

# Create second metric file
metric_file_2 = os.path.join(temp_dir, "metrics_2.json")
with open(metric_file_2, 'w') as f:
    json.dump({"precision": 0.92, "recall": 0.89}, f)

# Create third metric file
metric_file_3 = os.path.join(temp_dir, "metrics_3.json")
with open(metric_file_3, 'w') as f:
    json.dump({"f1_score": 0.90}, f)

ARGS = {
    "metric_files": [metric_file_1, metric_file_2, metric_file_3]
}
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 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
setup_logging Ran
ran; our contract draft was wrong, not the code
NVIDIA/NeMo-Skills/nemo_skills/utils.py
code served (permissive licence) · get_code("4261064eb4ec7756")
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 Logger. 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 logging
import sys

ARGS = {
    "disable_hydra_logs": True,
    "log_level": logging.INFO,
    "use_rich": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
rich — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Logger — a Logger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Logger 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
count_newlines Not yet run
unverified (harvested, no recorded run)
NVIDIA/NeMo-Skills/nemo_skills/file_utils.py
code served (permissive licence) · get_code("5b3c40e3214faf87")
How it ran

no invocation recorded.

load_config Not yet run
unverified (harvested, no recorded run)
NVIDIA/NeMo-Skills/nemo_skills/conversion/hf_to_nemo_llama.py
code served (permissive licence) · get_code("e74354657e62b51d")
How it ran

no invocation recorded.

load_config Not yet run
unverified (harvested, no recorded run)
NVIDIA/NeMo-Skills/nemo_skills/conversion/hf_to_nemo_qwen.py
code served (permissive licence) · get_code("de5cce096268a65e")
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, 15 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 15 later) and 0 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
calculate_chunk_indices 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
count_newlines 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
determine_proof_status 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
extract_code_output 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
extract_code_to_execute 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
extract_proof_only 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
format_code_output 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
get_logger_name 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
import_score_module 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
init_wandb 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
jload 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
load_config 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
load_config 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
load_metric_files 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA
setup_logging 2 other papers
2 later
2504.16891 (2025-04)
AIMO-2 Winning Solution: Building State-of-the-Art Mathematical Reasoning Models with OpenMathReasoning dataset
2604.12374 (2026-04)
Nemotron 3 Super: Open, Efficient Mixture-of-Experts Hybrid Mamba-Transformer Model for Agentic Reasoning NVIDIA

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

Mathematical reasoning continues to be a critical challenge in large language model (LLM) development with significant interest. However, most of the cutting-edge progress in mathematical reasoning with LLMs has become \emph{closed-source} due to lack of access to training data. This lack of data access limits researchers from understanding the impact of different choices for synthesizing and utilizing the data. With the goal of creating a high-quality finetuning (SFT) dataset for math reasoning, we conduct careful ablation experiments on data synthesis using the recently released \texttt{Llama3.1} family of models. Our experiments show that: (a) solution format matters, with excessively verbose solutions proving detrimental to SFT performance, (b) data generated by a strong teacher outperforms equally-sized data generated by a weak student model, (c) SFT is robust to low-quality solutions, allowing for imprecise data filtering, and (d) question diversity is crucial for achieving data scaling gains. Based on these insights, we create the OpenMathInstruct-2 dataset, which consists of 14M question-solution pairs ($\approx$ 600K unique questions), making it nearly eight times larger than the previous largest open-source math reasoning dataset. Finetuning the \texttt{Llama-3.1-8B-Base} using OpenMathInstruct-2 outperforms \texttt{Llama3.1-8B-Instruct} on MATH by an absolute 15.9\% (51.9\% $\rightarrow$ 67.8\%). Finally, to accelerate the open-source efforts, we release the code, the finetuned models, and the OpenMathInstruct-2 dataset under a commercially permissive license.

For agents

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

get_harvested_code_for_paper("2410.01560")
get_code_for_paper("2410.01560")
have("2410.01560")

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

curl https://syntology.ai/api/ran/2410.01560.json

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

[![Syntology run record](https://syntology.ai/api/ran/2410.01560.svg)](https://syntology.ai/paper/2410.01560)

Connect an agent — have() is free.