SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2403.04706 · 2024

Common 7B Language Models Already Possess Strong Math Capabilities

arXiv · PDF · Open in the Atlas

Code that ran

21 of 28 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; 0 honoured the contract we drafted, 0 violated it and 11 have no contract verdict; 10 failures of our instrument; 7 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/2403.04706.json · badge: /api/ran/2403.04706.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
xwin-lm/xwin-lm canonical (classifier label)
classifier basis: census
9 ran · 3 no recorded run
jerrywu-code/susgen community (classifier label)
classifier basis not recorded on the link
12 ran · 4 no recorded run
FunctionStatusWhere it lives
documentMetadata_loader Ran
ran on a synthesized input
jerrywu-code/susgen/utils/extractText_sus.py
code served (permissive licence) · get_code("ce0b6e5c855875fc")
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 os
import tempfile
import shutil

# Create a temporary directory structure with test files
temp_dir = tempfile.mkdtemp()

# Create subdirectories and files matching the expected format
# Format: <companyName>_<year>_<documentType>.<extension>

os.makedirs(os.path.join(temp_dir, 'subdir1'), exist_ok=True)

# Create valid test files
with open(os.path.join(temp_dir, 'CompanyA_2023_AnnualReport.pdf'), 'w') as f:
    f.write('test content')

with open(os.path.join(temp_dir, 'CompanyA_2022_QuarterlyReport.txt'), 'w') as f:
    f.write('test content')

with open(os.path.join(temp_dir, 'subdir1', 'CompanyB_2023_Disclosure.html'), 'w') as f:
    f.write('test content')

with open(os.path.join(temp_dir, 'CompanyC_2021_Statement.xhtml'), 'w') as f:
    f.write('test content')

# Create an invalid file that should be warned about
with open(os.path.join(temp_dir, 'InvalidFile.xlsx'), 'w') as f:
    f.write('test content')

ARGS = {
    'folderPath': temp_dir
}
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
extract_numbers Ran
ran on a synthesized input
jerrywu-code/susgen/eval/code/eval_finqa.py
code served (permissive licence) · get_code("72dfc5e9d73fa763")
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 json
import sys
import os
import re

# Set up random seed for reproducibility
import torch
torch.manual_seed(0)

# The function extract_numbers takes three parameters:
# 1. true_value: a string representing a number or percentage
# 2. predicted_text: a string containing text with numbers
# 3. threshold: a float (default 0.01)

# Let's create valid test cases
true_value_1 = "50%"
predicted_text_1 = "The result is approximately 0.5 or 50%"
threshold_1 = 0.01

ARGS = {
    "true_value": true_value_1,
    "predicted_text": predicted_text_1,
    "threshold": threshold_1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_text Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/HumanEval/humaneval_gen.py
pointer only (licence: NONE) · get_code("7a733b18081436f3")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

find_indentation Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/APPS/reindent.py
pointer only (licence: NONE) · get_code("f05983c38acfa250")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

generate_prompt Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/DS1000/ds1000_gen.py
pointer only (licence: NONE) · get_code("2763cff93674bd85")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

generate_prompt Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/HumanEval/humaneval_gen.py
pointer only (licence: NONE) · get_code("34de3c98b4c4ccf0")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

generate_text Ran
ran on a synthesized input
jerrywu-code/susgen/src/template.py
code served (permissive licence) · get_code("aa963276f597a210")
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

torch.manual_seed(0)

# Create a minimal tokenizer stand-in
class MinimalTokenizer:
    def encode(self, text, return_tensors=None):
        # Return a small tensor of token ids
        tokens = torch.tensor([[1, 2, 3, 4]], dtype=torch.long)
        if return_tensors == 'pt':
            return tokens
        return tokens
    
    def batch_decode(self, token_ids, skip_special_tokens=False):
        # Return decoded text with the expected format
        return ["### Response:\nThis is the answer"]

# Create a minimal model
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 8)
    
    def generate(self, input_ids, repetition_penalty=1.0, **kwargs):
        # Return token ids in the expected format
        # Must include "### Response:" in the output
        batch_size = input_ids.shape[0]
        # Return a tensor that will decode to text with "### Response:"
        return torch.tensor([[1, 2, 3, 4, 5, 6]], dtype=torch.long)

# Instantiate components
device = 'cpu'
model = MinimalModel()
tokenizer = MinimalTokenizer()
prompt = "What is the answer?"
args = {'max_length': 50, 'num_beams': 1}

ARGS = {
    'model': model,
    'tokenizer': tokenizer,
    'device': device,
    'prompt': prompt,
    'args': args
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 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
get_question Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/APPS/testing_util.py
pointer only (licence: NONE) · get_code("ed8331339b4de97c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_solutions Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/APPS/testing_util.py
pointer only (licence: NONE) · get_code("0c83bb0f43b8dc57")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_valid_problems Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/APPS/testing_util.py
pointer only (licence: NONE) · get_code("8bbb265e59e23293")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

import_source_file Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/DS1000/ds1000ds.py
pointer only (licence: NONE) · get_code("4557da831202481d")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

instr_prompt Ran
ran on a synthesized input
jerrywu-code/susgen/src/template.py
code served (permissive licence) · get_code("cd3318f643f2b529")
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

torch.manual_seed(0)

ARGS = {
    'content': 'This is a test instruction prompt'
}
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
llama3_short Ran
ran on a synthesized input
jerrywu-code/susgen/src/utils/prompt_template.py
code served (permissive licence) · get_code("836aab9d5828036c")
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

def llama3_short(record):
    if not record['input']:
        text = '### Instruction: {0}\n\n### Response: {1}<|eot_id|>'.format(record['instruction'], record['output'])
    else:
        text = '### Instruction: {0}\n\n### Input: {1}\n\n### Response: {2}<|eot_id|>'.format(record['instruction'], record['input'], record['output'])
    return text

# Construct a valid record dict
record = {
    'instruction': 'What is the capital of France?',
    'input': '',
    'output': 'The capital of France is Paris.'
}

ARGS = {
    'record': record
}
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
load_json Ran
ran; our contract draft was wrong, not the code
jerrywu-code/susgen/src/finetune.py
code served (permissive licence) · get_code("b3cd6beeda0e0909")
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 torch
import json
import tempfile
import os

torch.manual_seed(0)

# Create a temporary JSON file with some test data
temp_dir = tempfile.mkdtemp()
test_file = os.path.join(temp_dir, "test.json")
test_data = {
    "key1": "value1",
    "key2": [1, 2, 3],
    "key3": {"nested": "data"}
}
with open(test_file, 'w') as f:
    json.dump(test_data, f)

ARGS = {
    "file_path": test_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers, wandb, yaml — 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 on a synthesized input
jerrywu-code/susgen/utils/process_v1.py
code served (permissive licence) · get_code("991cad3f55659f76")
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 json
import tempfile
import os

# Create a temporary JSONL file with test data
temp_dir = tempfile.mkdtemp()
test_file = os.path.join(temp_dir, "test.jsonl")

# Write test JSONL data
test_data = [
    {"id": 1, "text": "hello world"},
    {"id": 2, "text": "foo bar"},
    {"id": 3, "name": "test", "value": 42}
]

with open(test_file, 'w', encoding='utf-8') as f:
    for item in test_data:
        f.write(json.dumps(item) + '\n')

ARGS = {
    "file": test_file
}
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_lora Ran
ran on a synthesized input
jerrywu-code/susgen/src/lora_inference.py
code served (permissive licence) · get_code("5acbb99c75b740a1")
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 MagicMock. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from unittest.mock import Mock, MagicMock
from peft import PeftModel

torch.manual_seed(0)

# Create a minimal base model (nn.Module)
class MinimalModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

base_model = MinimalModel()

# Create a mock lora_path (string path)
lora_path = "/tmp/mock_lora_model"

# Mock PeftModel.from_pretrained to avoid actual file I/O
original_from_pretrained = PeftModel.from_pretrained

def mock_from_pretrained(model, model_id, **kwargs):
    mock_model = MagicMock(spec=PeftModel)
    mock_model.model = model
    return mock_model

PeftModel.from_pretrained = staticmethod(mock_from_pretrained)

ARGS = {
    "base_model": base_model,
    "lora_path": lora_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
peft, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MagicMock — a MagicMock output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a MagicMock 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
mistral_formal Ran
ran on a synthesized input
jerrywu-code/susgen/src/utils/prompt_template.py
code served (permissive licence) · get_code("c09691f510dbbb2a")
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

# Construct valid arguments for mistral_formal function
record = {
    'input': '',
    'instruction': 'What is the capital of France?',
    'output': 'The capital of France is Paris.'
}

ARGS = {
    'record': record
}
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
mistral_short Ran
ran on a synthesized input
jerrywu-code/susgen/src/utils/prompt_template.py
code served (permissive licence) · get_code("46788756dee21d65")
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

torch.manual_seed(0)

# The function takes a 'record' parameter which is a dict
# It accesses: record['input'], record['instruction'], record['output']
# These are all strings based on how they're used in .format()

ARGS = {
    'record': {
        'instruction': 'What is the capital of France?',
        'input': '',
        'output': 'The capital of France is Paris.'
    }
}
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
replace_inline_tabs Ran
ran; our contract draft was wrong, not the code
xwin-lm/xwin-lm/Xwin-Coder/APPS/reindent.py
pointer only (licence: NONE) · get_code("27f23c533d62061a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

time_logger Ran
ran on a synthesized input
jerrywu-code/susgen/src/batch_inference.py
code served (permissive licence) · get_code("1efc261c7054e106")
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 torch.nn as nn
import time

torch.manual_seed(0)

def time_logger(func):
    def wrapper(*args, **kwargs):
        start_time = time.time()
        result = func(*args, **kwargs)
        end_time = time.time()
        elapsed_time = end_time - start_time
        print(f"Function '{func.__name__}' took {elapsed_time:.2f} seconds to execute.")
        return result
    return wrapper

# Example function to decorate
@time_logger
def sample_function(x, y):
    time.sleep(0.01)
    return x + y

# Arguments for time_logger
# time_logger takes one parameter: func (the function to be decorated)
ARGS = {
    "func": sample_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
tokenize Ran
ran on a synthesized input
jerrywu-code/susgen/src/finetune.py
code served (permissive licence) · get_code("448dfa384a102a35")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family int, shape (1, 6) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class MockTokenizer:
    """Minimal tokenizer that implements __call__ with text parameter"""
    def __call__(self, text, **kwargs):
        # Return a dict with 'input_ids' key containing token IDs
        # Simulate tokenization by converting text length to token count
        token_count = len(text.split())
        return {
            'input_ids': torch.tensor([[101] + [1000 + i for i in range(token_count)]])
        }

tokenizer = MockTokenizer()

hparams = {
    'return_tensors': 'pt',
    'padding': True,
    'truncation': True,
    'max_length': 128
}

prompt = 'This is a test sentence.'

ARGS = {
    'tokenizer': tokenizer,
    'hparams': hparams,
    'prompt': prompt
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family int, shape (1, 6) — shape (1, 6) 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
change_format Not yet run
unverified (harvested, no recorded run)
xwin-lm/xwin-lm/Xwin-Coder/DS1000/ds1000_gen.py
pointer only (licence: NONE) · get_code("b26735c9048e647a")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

extract_answer Not yet run
unverified (harvested, no recorded run)
jerrywu-code/susgen/eval/code/eval_finqa.py
code served (permissive licence) · get_code("30e951ff25cd61d7")
How it ran

no invocation recorded.

get_model Not yet run
unverified (harvested, no recorded run)
xwin-lm/xwin-lm/Xwin-Coder/HumanEval/humaneval_gen.py
pointer only (licence: NONE) · get_code("15a54731c6978e32")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

get_model Not yet run
unverified (harvested, no recorded run)
xwin-lm/xwin-lm/Xwin-Coder/MBPP/mbpp_gen.py
pointer only (licence: NONE) · get_code("54ff8bf66412b359")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, xwin-lm/xwin-lm when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

inference Not yet run
unverified (harvested, no recorded run)
jerrywu-code/susgen/src/lora_inference.py
code served (permissive licence) · get_code("5adc840cfa3cba8a")
How it ran

no invocation recorded.

load_datasets Not yet run
unverified (harvested, no recorded run)
jerrywu-code/susgen/utils/loadhf_data.py
code served (permissive licence) · get_code("43fd4e86eac55a80")
How it ran

no invocation recorded.

load_model Not yet run
unverified (harvested, no recorded run)
jerrywu-code/susgen/src/template.py
code served (permissive licence) · get_code("c87d1b4488a54c82")
How it ran

no invocation recorded.

Also present in other papers' code

Of 28 samples lifted from this paper's linked repositories, 28 carry a function identity and 0 do not. Of the 28 distinct functions with an identity, 7 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 0 the same month, 4 later) and 21 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_solutions 4 other papers
3 dated earlier, 1 later
2105.09938 (2021-05)
Measuring Coding Challenge Competence With APPS
2107.03374 (2021-07)
Evaluating Large Language Models Trained on Code
2203.08597 (2022-03)
Less is More: Summary of Long Instructions is Better for Program Synthesis
and 1 more not named here
get_valid_problems 4 other papers
3 dated earlier, 1 later
2105.09938 (2021-05)
Measuring Coding Challenge Competence With APPS
2107.03374 (2021-07)
Evaluating Large Language Models Trained on Code
2203.08597 (2022-03)
Less is More: Summary of Long Instructions is Better for Program Synthesis
and 1 more not named here
extract_text 3 other papers
3 dated earlier
2305.07922 (2023-05)
CodeT5+: Open Code Large Language Models for Code Understanding and Generation
2306.08568 (2023-06)
WizardCoder: EMPOWERING CODE LARGE LAN-GUAGE MODELS WITH EVOL-INSTRUCT
2308.09583 (2023-08)
WizardMath: EMPOWERING MATHEMATICAL REASON-ING FOR LARGE LANGUAGE MODELS VIA Reinforced Evol-Instruct
get_question 3 other papers
2 dated earlier, 1 later
2105.09938 (2021-05)
Measuring Coding Challenge Competence With APPS
2107.03374 (2021-07)
Evaluating Large Language Models Trained on Code
2602.01685 (2026-02)
Semantic-aware Wasserstein Policy Regularization for Large Language Model Alignment
find_indentation 2 other papers
2 dated earlier
2105.09938 (2021-05)
Measuring Coding Challenge Competence With APPS
2107.03374 (2021-07)
Evaluating Large Language Models Trained on Code
replace_inline_tabs 2 other papers
2 dated earlier
2105.09938 (2021-05)
Measuring Coding Challenge Competence With APPS
2107.03374 (2021-07)
Evaluating Large Language Models Trained on Code
load_json 1 other paper
1 later
2412.06660 (2024-12)
MuMu-LLaMA: Multi-modal Music Understanding and Generation via Large Language Models

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 capabilities were previously believed to emerge in common language models only at a very large scale or require extensive math-related pre-training. This paper shows that the LLaMA-2 7B model with common pre-training already exhibits strong mathematical abilities, as evidenced by its impressive accuracy of 97.7% and 72.0% on the GSM8K and MATH benchmarks, respectively, when selecting the best response from 256 random generations. The primary issue with the current base model is the difficulty in consistently eliciting its inherent mathematical capabilities. Notably, the accuracy for the first answer drops to 49.5% and 7.9% on the GSM8K and MATH benchmarks, respectively. We find that simply scaling up the SFT data can significantly enhance the reliability of generating correct answers. However, the potential for extensive scaling is constrained by the scarcity of publicly available math questions. To overcome this limitation, we employ synthetic data, which proves to be nearly as effective as real data and shows no clear saturation when scaled up to approximately one million samples. This straightforward approach achieves an accuracy of 82.6% on GSM8K and 40.6% on MATH using LLaMA-2 7B models, surpassing previous models by 14.2% and 20.8%, respectively. We also provide insights into scaling behaviors across different reasoning complexities and error types.

For agents

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

get_harvested_code_for_paper("2403.04706")
get_code_for_paper("2403.04706")
have("2403.04706")

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

curl https://syntology.ai/api/ran/2403.04706.json

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

[![Syntology run record](https://syntology.ai/api/ran/2403.04706.svg)](https://syntology.ai/paper/2403.04706)

Connect an agent — have() is free.