SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2407.13690 · NeurIPS · 2024

DART-Math: Difficulty-Aware Rejection Tuning for Mathematical Problem-Solving

Junxian He, Rui Wang, Xiwen Zhang, Yuxuan Tong, Ruidong Wu

arXiv · PDF · Open in the Atlas

Code that ran

10 of 12 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; 9 failures of our instrument; 2 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/2407.13690.json · badge: /api/ran/2407.13690.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
hkust-nlp/dart-math canonical (classifier label)
classifier basis: census
10 ran · 2 no recorded run
FunctionStatusWhere it lives
async_wrap Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/parallel.py
code served (permissive licence) · get_code("ace871f6d9931df1")
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 function. 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 functools
from typing import Any, Awaitable, Callable

# Define a simple synchronous function to wrap
def simple_function(x: int, y: str = "default") -> str:
    """A simple synchronous function for testing."""
    return f"{y}: {x}"

# The function to wrap
def async_wrap(func: Callable) -> Awaitable:
    """Wrap a synchronous function `func` into an asynchronous function."""

    @functools.wraps(func)
    async def coroutine_wrapper(*args, **kwargs):
        return func(*args, **kwargs)
    return coroutine_wrapper

# Construct the ARGS dictionary
ARGS = {
    "func": simple_function
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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_ans_from_math_sol Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/data.py
code served (permissive licence) · get_code("f184066eadcee200")
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

# This function takes a single string parameter and extracts boxed content from it
# Let's create valid test cases

# Case 1: Valid boxed answer
valid_boxed_string = r"Here is the solution: \boxed{42}"

# Case 2: Valid fbox answer
valid_fbox_string = r"Here is the solution: \fbox{3.14}"

# Case 3: Nested braces
nested_braces_string = r"Solution: \boxed{x = \{1, 2, 3\}}"

# Case 4: No boxed/fbox - should return None
no_boxed_string = "This is just a regular string with no answer"

# Case 5: Incomplete boxed - should return None
incomplete_boxed_string = r"Solution: \boxed{incomplete"

# Use the valid boxed case as the main test case
ARGS = {
    "string": valid_boxed_string
}
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
extract_level_from_math_dp Ran
ran, honoured the contract we drafted
hkust-nlp/dart-math/dart_math/data.py
code served (permissive licence) · get_code("4abb06623d7c704d")
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, 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
import numpy as np

# The function expects a dictionary 'dp' with keys 'level' and 'problem'
# 'level' is a string that can be split by space, with the last element being either a digit or '?'
# 'problem' is a string that may start with 'We'

ARGS = {
    'dp': {
        'level': 'Grade 5',
        'problem': 'We need to solve this math problem.'
    }
}
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 int, ndim 0, dtype family int, 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
get_answer_type_text Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/olympiadbench.py
code served (permissive licence) · get_code("bed2810b31296fa1")
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

# Analyze the function signature and behavior
# get_answer_type_text(answer_type, is_chinese, multiple_answer)
# 
# Parameters:
# - answer_type: str, used with 'in' checks and split operations
# - is_chinese: bool, used in if-else conditionals
# - multiple_answer: bool, used in if-else conditionals

# The function checks:
# - 'Need_human_evaluate' in answer_type
# - 'Tuple' in answer_type
# - '-' in answer_type (processed via find)
# - ',' in answer_type (split by comma)
# - calls get_single_answer_type_text which looks for ['Numerical', 'Expression', 'Equation', 'Interval']

# Test case 1: Simple case with single answer type, English, not multiple
answer_type_1 = "Numerical"
is_chinese_1 = False
multiple_answer_1 = False

# Test case 2: Multiple answers with comma-separated types, Chinese
answer_type_2 = "Expression,Equation"
is_chinese_2 = True
multiple_answer_2 = True

# Test case 3: Need_human_evaluate (should return empty string)
answer_type_3 = "Need_human_evaluate"
is_chinese_3 = False
multiple_answer_3 = False

# Test case 4: Multiple answers same type, English
answer_type_4 = "Interval"
is_chinese_4 = False
multiple_answer_4 = True

# Test case 5: Answer type with dash prefix, Chinese
answer_type_5 = "Numerical-some_suffix"
is_chinese_5 = True
multiple_answer_5 = False

# Using test case 1 as the primary example
ARGS = {
    'answer_type': 'Numerical',
    'is_chinese': False,
    'multiple_answer': False
}
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
get_pathname_from_name_or_path Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/utils.py
code served (permissive licence) · get_code("f249e05f5c701d28")
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 logging
import os
import tempfile

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

# Test case 1: HuggingFace Hub style name (no local path exists)
hf_name = "bert-base-uncased"

# Test case 2: Local file path
local_file = os.path.join(temp_dir, "model.bin")
with open(local_file, 'w') as f:
    f.write("test")

# Test case 3: Local directory with checkpoint
checkpoint_dir = os.path.join(temp_dir, "checkpoint-1000")
os.makedirs(checkpoint_dir, exist_ok=True)

# Test case 4: Absolute path
absolute_path = os.path.join(temp_dir, "mymodel")
os.makedirs(absolute_path, exist_ok=True)

# We'll use a simple HF-style name as the primary test case
ARGS = {
    "name_or_path": "bert-base-uncased"
}
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
get_single_answer_type_text Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/olympiadbench.py
code served (permissive licence) · get_code("0047d2388e36a9dd")
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 = {
    'answer_type': 'Numerical-something',
    'is_chinese': True
}
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
load_json Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/utils.py
code served (permissive licence) · get_code("3c40f841274b3441")
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 orjson
import tempfile
import os

# Create a temporary JSON file for testing
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test.json")

# Write a valid JSON file
test_data = {"key1": "value1", "key2": [1, 2, 3], "key3": {"nested": True}}
with open(temp_file, 'w') as f:
    f.write(orjson.dumps(test_data).decode('utf-8'))

ARGS = {
    "fpath": temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
orjson — 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
load_jsonl Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/utils.py
code served (permissive licence) · get_code("fb57a712227893bd")
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 tempfile
import orjson

# Create a temporary JSONL file
temp_file = tempfile.NamedTemporaryFile(mode='w', suffix='.jsonl', delete=False)
temp_fpath = temp_file.name

# Write some test data to the JSONL file
test_data = [
    {'id': 1, 'text': 'hello world'},
    {'id': 2, 'text': 'foo bar'},
    {'id': 3, 'text': 'test data'},
]

for item in test_data:
    temp_file.write(orjson.dumps(item).decode('utf-8') + '\n')

temp_file.close()

ARGS = {
    'fpath': temp_fpath,
    'use_tqdm': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
orjson, tqdm — 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
make_prompt Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/olympiadbench.py
code served (permissive licence) · get_code("e6eeb93f5a059925")
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

# Construct a question dictionary with all required fields
question_chinese_math = {
    'answer_type': 'Numerical',
    'is_multiple_answer': False,
    'unit': 'm',
}

question_chinese_math_multiple = {
    'answer_type': 'Expression,Equation',
    'is_multiple_answer': True,
    'unit': '',
}

question_english_physics = {
    'answer_type': 'Interval',
    'is_multiple_answer': False,
    'unit': 'N',
}

question_english_physics_multiple = {
    'answer_type': 'Numerical',
    'is_multiple_answer': True,
    'unit': '',
}

question_need_human = {
    'answer_type': 'Need_human_evaluate',
    'is_multiple_answer': False,
    'unit': '',
}

question_tuple = {
    'answer_type': 'Tuple',
    'is_multiple_answer': False,
    'unit': '',
}

# Test case 1: Chinese math problem, single answer with unit
ARGS = {
    'question': question_chinese_math,
    'is_chinese': True,
    'is_math': True,
    'is_theorem_proving': False,
}
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
seq_consume_preset_queue_w_each_timeout Ran
ran; our contract draft was wrong, not the code
hkust-nlp/dart-math/dart_math/parallel.py
code served (permissive licence) · get_code("c2dd672c3bad60d7")
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 coroutine. 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 asyncio
import queue
from tqdm import tqdm


async def dummy_consumer(task_id=None, value=None):
    """A simple async consumer function for testing."""
    await asyncio.sleep(0.01)
    return f"processed_{task_id}_{value}"


def sync_consumer(task_id=None, value=None):
    """A simple sync consumer function for testing."""
    return f"processed_{task_id}_{value}"


# Create a queue with indexed kwargs
idxed_kwargs_queue = queue.SimpleQueue()
idxed_kwargs_queue.put((0, {"task_id": 0, "value": 10}))
idxed_kwargs_queue.put((1, {"task_id": 1, "value": 20}))
idxed_kwargs_queue.put((2, {"task_id": 2, "value": 30}))
idxed_kwargs_queue.put(None)  # Sentinel to mark end of queue

# Create a progress bar
pbar = tqdm(total=3, desc="Processing tasks")

ARGS = {
    "consumer": dummy_consumer,
    "idxed_kwargs_queue": idxed_kwargs_queue,
    "timeout": 10,
    "pbar": pbar,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type coroutine — a coroutine output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a coroutine 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_icl_egs Not yet run
unverified (harvested, no recorded run)
hkust-nlp/dart-math/dart_math/gen.py
code served (permissive licence) · get_code("ecd6408282b087da")
How it ran

no invocation recorded.

get_n_shots Not yet run
unverified (harvested, no recorded run)
hkust-nlp/dart-math/dart_math/gen.py
code served (permissive licence) · get_code("ac56cd1c0989eb82")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 12 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 12 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
get_answer_type_text 2 other papers
1 dated earlier, 1 later
2402.14008 (2024-02)
OlympiadBench: A Challenging Benchmark for Promoting AGI with Olympiad-Level Bilingual Multimodal Scientific Problems
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
async_wrap 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
extract_ans_from_math_sol 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
extract_level_from_math_dp 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
get_icl_egs 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
get_n_shots 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
get_pathname_from_name_or_path 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
get_single_answer_type_text 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
load_json 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
load_jsonl 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
make_prompt 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion
seq_consume_preset_queue_w_each_timeout 1 other paper
1 later
2503.16212 (2025-03)
MathFusion: Enhancing Mathematical Problem-solving of LLM through Instruction Fusion

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

Solving mathematical problems requires advanced reasoning abilities and presents notable challenges for large language models. Previous works usually synthesize data from proprietary models to augment existing datasets, followed by instruction tuning to achieve top-tier results. However, our analysis of these datasets reveals severe biases towards easy queries, with frequent failures to generate any correct response for the most challenging queries. Hypothesizing that difficult queries are crucial to learning complex reasoning, we propose Difficulty-Aware Rejection Tuning (DART), a method that allocates difficult queries more trials during the synthesis phase, enabling more extensive training on difficult samples. Utilizing DART, we have created new datasets for mathematical problem-solving that focus more on difficult queries and are substantially smaller than previous ones. Remarkably, our synthesis process solely relies on a 7B-sized open-weight model, without reliance on the commonly used proprietary GPT-4. We fine-tune various base models on our datasets ranging from 7B to 70B in size, resulting in a series of strong models called DART-Math. In comprehensive in-domain and out-of-domain evaluation on 6 mathematical benchmarks, DART-Math outperforms vanilla rejection tuning significantly, being superior or comparable to previous arts, despite using much smaller datasets and no proprietary models. Furthermore, our results position our synthetic datasets as the most effective and cost-efficient publicly available resources for advancing mathematical problem-solving. 1 * Work done during visit to HKUST. 1 Our datasets, models and code are publicly available at https://github.com/hkust-nlp/dart-math. 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("2407.13690")
get_code_for_paper("2407.13690")
have("2407.13690")

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

curl https://syntology.ai/api/ran/2407.13690.json

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

[![Syntology run record](https://syntology.ai/api/ran/2407.13690.svg)](https://syntology.ai/paper/2407.13690)

Connect an agent — have() is free.