SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.03300 · 2024

DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models

Junxiao Song, Daya Guo, Peiyi Wang, Zhihong Shao, Y Li, Runxin Xu, Haowei Zhang, Xiao Bi, Y Wu, Mingchuan Zhang, Qihao Zhu

arXiv · PDF · Open in the Atlas

Code that ran

14 of 24 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 2 have no contract verdict; 12 failures of our instrument; 10 no recorded run); 1 of them was also run on the host, not in the jail, by the peer-agreement lane that set its level, 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/2402.03300.json · badge: /api/ran/2402.03300.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
deepseek-ai/deepseek-math canonical (classifier label)
classifier basis: census
8 ran · 1 no recorded run
yynil/rwkvinside community (classifier label)
classifier basis not recorded on the link
5 ran · 9 no recorded run
openpsi-project/realhf community
Papers with Code archive
1 ran · 0 no recorded run
FunctionStatusWhere it lives
repeat_kv Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
yynil/rwkvinside/rwkv_inside/modeling_arwkv.py
code served (permissive licence) · get_code("30d7eec482ebf6b1")
How it ran

Ran in Syntology's jail on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 4, dtype family float, shape (2, 6, 5, 4) as a later run recorded it. The image it ran in was not recorded. The comparison behind this sample's level is a different run: the peer-agreement lane ran it again on the host, not in the jail (report dated 2026-09-04T23:09:20Z). If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a tensor with shape (batch, num_key_value_heads, seqlen, head_dim)
# Using distinct sizes: batch=2, num_key_value_heads=3, seqlen=5, head_dim=4
hidden_states = torch.randn(2, 3, 5, 4)

# n_rep is the repetition factor
n_rep = 2

ARGS = {
    "hidden_states": hidden_states,
    "n_rep": n_rep,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 4, dtype family float, shape (2, 6, 5, 4) — shape (2, 6, 5, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
actor_loss_fn Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
openpsi-project/realhf/realhf/impl/model/utils/ppo_functional.py
code served (permissive licence) · get_code("e2e940c65c7460d3")
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) 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

torch.manual_seed(0)

# Create small tensors with matching shapes
batch_size = 2
seq_len = 4

logprobs = torch.randn(batch_size, seq_len, dtype=torch.float32)
old_logprobs = torch.randn(batch_size, seq_len, dtype=torch.float32)
advantages = torch.randn(batch_size, seq_len, dtype=torch.float32)

# Create a loss mask (optional, but we'll include it)
loss_mask = torch.ones(batch_size, seq_len, dtype=torch.bool)
loss_mask[0, 2] = False  # Mark one position as invalid

eps_clip = 0.2

ARGS = {
    "logprobs": logprobs,
    "old_logprobs": old_logprobs,
    "advantages": advantages,
    "eps_clip": eps_clip,
    "loss_mask": loss_mask,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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
aggregate_metrics Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/summarize_results.py
code served (permissive licence) · get_code("8b9e2baf73bb9a6f")
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 directory and metric files
temp_dir = tempfile.mkdtemp()

# Create first metric file
metric1 = {
    'n_samples': 100,
    'accuracy': 0.95,
    'loss': 0.05
}
path1 = os.path.join(temp_dir, 'metric1.json')
with open(path1, 'w') as f:
    json.dump(metric1, f)

# Create second metric file
metric2 = {
    'n_samples': 50,
    'accuracy': 0.90,
    'loss': 0.10
}
path2 = os.path.join(temp_dir, 'metric2.json')
with open(path2, 'w') as f:
    json.dump(metric2, f)

# Define ARGS
ARGS = {
    'paths': [path1, path2]
}
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
aggregate_predictions Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/summarize_results.py
code served (permissive licence) · get_code("33425041e7f88b7f")
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 temporary JSON files for testing
temp_dir = tempfile.mkdtemp()

# Create valid JSON file 1
path1 = os.path.join(temp_dir, "predictions1.json")
with open(path1, 'w') as f:
    json.dump([{"id": 1, "prediction": 0.9}, {"id": 2, "prediction": 0.7}], f)

# Create valid JSON file 2
path2 = os.path.join(temp_dir, "predictions2.json")
with open(path2, 'w') as f:
    json.dump([{"id": 3, "prediction": 0.8}], f)

# Create invalid JSON file (will be skipped due to try-except)
path3 = os.path.join(temp_dir, "predictions3.json")
with open(path3, 'w') as f:
    f.write("invalid json content {{{")

# Define arguments
ARGS = {
    "paths": [path1, path2, path3]
}
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
apply_rotary_pos_emb Ran
ran; our contract draft was wrong, not the code
yynil/rwkvinside/rwkv_inside/modeling_arwkv.py
code served (permissive licence) · get_code("bac65c3dafaec040")
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) and returned tuple, ndim 5, 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

torch.manual_seed(0)

# Create small tensors for query and key
# Shape: [batch_size, heads, seq_len, head_dim]
batch_size = 2
heads = 3
seq_len = 4
head_dim = 8

q = torch.randn(batch_size, heads, seq_len, head_dim)
k = torch.randn(batch_size, heads, seq_len, head_dim)

# Create cos and sin tensors
# Shape: [batch_size, seq_len, head_dim]
cos = torch.randn(batch_size, seq_len, head_dim)
sin = torch.randn(batch_size, seq_len, head_dim)

# position_ids is optional and unused according to docstring
position_ids = None

# unsqueeze_dim defaults to 1
unsqueeze_dim = 1

ARGS = {
    "q": q,
    "k": k,
    "cos": cos,
    "sin": sin,
    "position_ids": position_ids,
    "unsqueeze_dim": unsqueeze_dim,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 5, 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
compare_latex_numbers Ran
ran on a synthesized input
yynil/rwkvinside/rl/utilities.py
code served (permissive licence) · get_code("620dc092b02e33f7")
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 re
import math

def parse_coordinate(coord_str):
    """解析坐标字符串,返回(x, y)元组"""
    cleaned = re.sub('\\s+', '', coord_str)
    match = re.match('\\(\\s*([-+]?\\d*\\.?\\d+)\\s*,\\s*([-+]?\\d*\\.?\\d+)\\s*\\)', cleaned)
    if match:
        return (float(match.group(1)), float(match.group(2)))
    return None

def parse_latex_to_python(expr):
    if isinstance(expr, (int, float)):
        return float(expr)
    expr = re.sub('\\s+', '', expr)
    coord = parse_coordinate(expr)
    if coord is not None:
        return coord
    expr = expr.replace(',', '')
    expr = re.sub('\\\\frac{([^{}]+)}{([^{}]+)}', '(\\1)/(\\2)', expr)
    expr = re.sub('\\\\dfrac{([^{}]+)}{([^{}]+)}', '(\\1)/(\\2)', expr)
    expr = re.sub('\\\\sqrt{([^{}]+)}', 'math.sqrt(\\1)', expr)
    expr = re.sub('\\\\sqrt\\[(\\d+)\\]{([^{}]+)}', 'math.pow(\\2, 1/\\1)', expr)
    expr = re.sub('([a-zA-Z0-9]+)\\^([a-zA-Z0-9]+)', '\\1**\\2)', expr)
    expr = re.sub('\\\\sin{([^{}]+)}', 'math.sin(\\1)', expr)
    expr = re.sub('\\\\cos{([^{}]+)}', 'math.cos(\\1)', expr)
    expr = re.sub('\\\\tan{([^{}]+)}', 'math.tan(\\1)', expr)
    expr = re.sub('\\\\log\\{([^{}]+)\\}\\{([^{}]+)\\}', 'math.log(\\2, \\1)', expr)
    expr = re.sub('\\\\exp{([^{}]+)}', 'math.exp(\\1)', expr)
    expr = re.sub('\\\\ln{([^{}]+)}', 'math.log(\\1)', expr)
    expr = re.sub('\\\\text{([^{}]+)}', '\\1', expr)
    expr = re.sub('\\$([0-9.]+)', '\\1', expr)
    expr = re.sub('£([0-9.]+)', '\\1', expr)
    expr = re.sub('€([0-9.]+)', '\\1', expr)
    expr = re.sub('¥([0-9.]+)', '\\1', expr)
    expr = re.sub('฿([0-9.]+)', '\\1', expr)
    try:
        result = eval(expr)
        return float(result) if isinstance(result, (int, float)) else result
    except Exception:
        return expr.strip()

def compare_latex_numbers(expr1, expr2):
    parsed1 = parse_latex_to_python(expr1)
    parsed2 = parse_latex_to_python(expr2)
    if isinstance(parsed1, tuple) and isinstance(parsed2, tuple):
        return abs(parsed1[0] - parsed2[0]) < 1e-06 and abs(parsed1[1] - parsed2[1]) < 1e-06
    if isinstance(parsed1, (int, float)) and isinstance(parsed2, (int, float)):
        return abs(parsed1 - parsed2) < 1e-06
    return str(parsed1) == str(parsed2)

ARGS = {
    "expr1": "42.5",
    "expr2": "42.5"
}
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 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_boxed_answers Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/data_processing/answer_extraction.py
pointer only (licence: MIT); the stored copy get_code serves, from PKU-SEC-Lab/SPEX, is pointer only (licence: NONE) · get_code("6390dea763ce122e")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from PKU-SEC-Lab/SPEX, and GitHub detected no licence in that repository 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.

extract_program_output Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/data_processing/answer_extraction.py
pointer only (licence: MIT); the stored copy get_code serves, from PKU-SEC-Lab/SPEX, is pointer only (licence: NONE) · get_code("0772e5020f6b0849")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from PKU-SEC-Lab/SPEX, and GitHub detected no licence in that repository 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.

markup_question Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/run_subset_parallel.py
code served (permissive licence) · get_code("de0231790ab22287")
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

ARGS = {
    'args': None,
    'item': {
        'messages': [
            {'content': 'What is 2+2?'},
            {'content': 'The answer is 4.'},
            {'content': 'What is 3+3?'},
            {'content': 'The answer is 6.'},
            {'content': 'What is 5+5?'}
        ]
    },
    'language': 'en',
    'src': 'test_source',
    'task': 'cot'
}
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 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
parse_coordinate Ran
ran on a synthesized input
yynil/rwkvinside/rl/utilities.py
code served (permissive licence) · get_code("eb7f513a0e05ec8b")
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 re

def parse_coordinate(coord_str):
    """解析坐标字符串,返回(x, y)元组"""
    cleaned = re.sub('\\s+', '', coord_str)
    match = re.match('\\(\\s*([-+]?\\d*\\.?\\d+)\\s*,\\s*([-+]?\\d*\\.?\\d+)\\s*\\)', cleaned)
    if match:
        return (float(match.group(1)), float(match.group(2)))
    return None


ARGS = {
    "coord_str": "(3.14, -2.71)"
}
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 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
read_data Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/utils.py
code served (permissive licence) · get_code("3deb39de933b985a")
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 temporary JSON file
json_fd, json_path = tempfile.mkstemp(suffix='.json')
json_data = [{"key": "value1"}, {"key": "value2"}]
with os.fdopen(json_fd, 'w') as f:
    json.dump(json_data, f)

# Create temporary JSONL file
jsonl_fd, jsonl_path = tempfile.mkstemp(suffix='.jsonl')
with os.fdopen(jsonl_fd, 'w') as f:
    f.write(json.dumps({"item": "first"}) + '\n')
    f.write(json.dumps({"item": "second"}) + '\n')

ARGS = {
    "path": json_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 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
rotate_half Ran
ran; our contract draft was wrong, not the code
yynil/rwkvinside/rwkv_inside/modeling_arwkv.py
code served (permissive licence) · get_code("b99eea6376d1e212")
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) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function rotate_half takes a single parameter 'x' which is a tensor
# It accesses x.shape[-1] to get the last dimension
# It uses indexing with [..., :] slicing notation
# It calls torch.cat() to concatenate tensors

# Create a small tensor with shape (2, 4) - batch size 2, hidden dim 4
x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
shuffle Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/utils.py
code served (permissive licence) · get_code("1d4da55afb0377ec")
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 random
import numpy as np

# The shuffle function takes:
# 1. data - a list-like object that has len() and supports indexing
# 2. seed - an integer

# Create sample data as a simple list
sample_data = ['item_0', 'item_1', 'item_2', 'item_3', 'item_4']

# Create ARGS dict with valid arguments
ARGS = {
    'data': sample_data,
    'seed': 42
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
strip_string Ran
ran; our contract draft was wrong, not the code
deepseek-ai/deepseek-math/evaluation/data_processing/answer_extraction.py
pointer only (licence: MIT); the stored copy get_code serves, from PKU-SEC-Lab/SPEX, is pointer only (licence: NONE) · get_code("5abed389274c4c03")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from PKU-SEC-Lab/SPEX, and GitHub detected no licence in that repository 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.

calculate_language_consistency_score Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/rl/deepspeed_rl_train.py
code served (permissive licence) · get_code("8ddbdfd0041abedc")
How it ran

no invocation recorded.

calculate_param_sizes Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/train_scripts/split_model_params.py
code served (permissive licence) · get_code("93678d5fdf25571d")
How it ran

no invocation recorded.

calculate_tensor_size Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/train_scripts/split_model_params.py
code served (permissive licence) · get_code("bc474830fb3ccd5d")
How it ran

no invocation recorded.

check_proof Not yet run
unverified (harvested, no recorded run)
deepseek-ai/deepseek-math/evaluation/unsafe_score_minif2f_isabelle.py
code served (permissive licence) · get_code("c59c51644b25a189")
How it ran

no invocation recorded.

distribute_params Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/train_scripts/split_model_params.py
code served (permissive licence) · get_code("d0a5e652d9881e59")
How it ran

no invocation recorded.

grpo_loss_with_old_logps Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/rl/grpo_trainer.py
code served (permissive licence) · get_code("f9c058ca4549e9f6")
How it ran

no invocation recorded.

parse_latex_to_python Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/rl/utilities.py
code served (permissive licence) · get_code("f7cc3a0c10540f2e")
How it ran

no invocation recorded.

reward_function Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/rl/train_rl_model.py
code served (permissive licence) · get_code("f872aa95062d3ac2")
How it ran

no invocation recorded.

selective_log_softmax_old Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/rl/grpo_trainer.py
code served (permissive licence) · get_code("7c05d945015e9e96")
How it ran

no invocation recorded.

time_function Not yet run
unverified (harvested, no recorded run)
yynil/rwkvinside/rl/profiler.py
code served (permissive licence) · get_code("b016e9b55d3c8dad")
How it ran

no invocation recorded.

Also present in other papers' code

Of 24 samples lifted from this paper's linked repositories, 24 carry a function identity and 0 do not. Of the 24 distinct functions with an identity, 12 are also present in other papers whose code we hold (3 with at least one paper dated earlier, 3 the same month, 12 later) and 12 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
rotate_half 360 other papers
68 dated earlier, 25 the same month, 258 later, 6 same year, order not determinable, 3 undated · 30 only as an assembly brick
1901.02860 (2019-01)
Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
2203.08913 (2022-03)
Memorizing Transformers
2302.13971 (2023-02)
LLaMA: Open and Efficient Foundation Language Models
and 357 more not named here
repeat_kv 243 other papers
33 dated earlier, 19 the same month, 187 later, 2 same year, order not determinable, 2 undated · 22 only as an assembly brick
1901.02860 (2019-01)
Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
2203.08913 (2022-03)
Memorizing Transformers
2305.17888 (2023-05)
LLM-QAT: Data-Free Quantization Aware Training for Large Language Models
and 240 more not named here
apply_rotary_pos_emb 94 other papers
4 dated earlier, 3 the same month, 83 later, 1 same year, order not determinable, 3 undated · 6 only as an assembly brick
1901.02860 (2019-01)
Transformer-XL: Attentive Language Models Beyond a Fixed-Length Context
2305.17888 (2023-05)
LLM-QAT: Data-Free Quantization Aware Training for Large Language Models
2309.11674 (2023-09)
A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models
and 91 more not named here
extract_boxed_answers 7 other papers
7 later
2406.18629 (2024-06)
Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2502.12067 (2025-02)
TokenSkip: Controllable Chain-of-Thought Compression in LLMs
and 4 more not named here
strip_string 6 other papers
6 later
2406.18629 (2024-06)
Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2502.12067 (2025-02)
TokenSkip: Controllable Chain-of-Thought Compression in LLMs
and 3 more not named here
extract_program_output 5 other papers
5 later
2406.18629 (2024-06)
Step-DPO: Step-wise Preference Optimization for Long-chain Reasoning of LLMs
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2502.12067 (2025-02)
TokenSkip: Controllable Chain-of-Thought Compression in LLMs
and 2 more not named here
read_data 3 other papers
3 later
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2502.12067 (2025-02)
TokenSkip: Controllable Chain-of-Thought Compression in LLMs
2506.20512 (2025-06)
OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
aggregate_metrics 2 other papers
2 later
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2506.20512 (2025-06)
OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
aggregate_predictions 2 other papers
2 later
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2506.20512 (2025-06)
OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
markup_question 2 other papers
2 later
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2506.20512 (2025-06)
OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
shuffle 2 other papers
2 later
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing
2506.20512 (2025-06)
OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
check_proof 1 other paper
1 later
2025.acl-long.1506 (2025 (year only))
Subtle Errors in Reasoning: Preference Learning via Error-injected Self-editing

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 poses a significant challenge for language models due to its complex and structured nature. In this paper, we introduce DeepSeekMath 7B, which continues pretraining DeepSeek-Coder-Base-v1.5 7B with 120B math-related tokens sourced from Common Crawl, together with natural language and code data. DeepSeekMath 7B has achieved an impressive score of 51.7% on the competition-level MATH benchmark without relying on external toolkits and voting techniques, approaching the performance level of Gemini-Ultra and GPT-4. Self-consistency over 64 samples from DeepSeekMath 7B achieves 60.9% on MATH. The mathematical reasoning capability of DeepSeekMath is attributed to two key factors: First, we harness the significant potential of publicly available web data through a meticulously engineered data selection pipeline. Second, we introduce Group Relative Policy Optimization (GRPO), a variant of Proximal Policy Optimization (PPO), that enhances mathematical reasoning abilities while concurrently optimizing the memory usage of PPO.

For agents

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

get_harvested_code_for_paper("2402.03300")
get_code_for_paper("2402.03300")
have("2402.03300")

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

curl https://syntology.ai/api/ran/2402.03300.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.03300.svg)](https://syntology.ai/paper/2402.03300)

Connect an agent — have() is free.