SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.00599 · 2024

EvoCodeBench: An Evolving Code Generation Benchmark Aligned with Real-World Code Repositories

arXiv · PDF · Open in the Atlas

Code that ran

15 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 6 honoured the contract we drafted, 3 violated it and 0 have no contract verdict; 6 failures of our instrument; 0 no recorded run), measured 2026-11-21; 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/2404.00599.json · badge: /api/ran/2404.00599.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
seketeam/evocodebench canonical (classifier label)
classifier basis: census
15 ran · 0 no recorded run
FunctionStatusWhere it lives
adjust_indent Ran
ran; our contract draft was wrong, not the code
seketeam/evocodebench/pass_k.py
code served (permissive licence) · get_code("020a90b56537e147")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) 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 textwrap

# Sample code with some indentation
code = """
def example_function():
    x = 1
    y = 2
    if x < y:
        print("x is less than y")
        return True
    return False
"""

# New indentation level (number of spaces)
new_indent = 8

ARGS = {
    "code": code,
    "new_indent": new_indent
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
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
compute_pass_at_k Ran
ran, honoured the contract we drafted
seketeam/evocodebench/pass_k.py
code served (permissive licence) · get_code("41f57204e706d18f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned float64, 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 numpy as np

# The function computes the pass@k metric for code generation evaluation
# n: total number of completions per task
# c: number of completions that pass all tests
# k: k in pass_at_k

# Example: 10 total completions, 3 passed, compute pass@1
n = 10
c = 3
k = 1

ARGS = {
    'n': n,
    'c': c,
    'k': k
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float64, 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
compute_recall Ran
ran, honoured the contract we drafted
seketeam/evocodebench/parser/recall_k.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from seketeam/deveval, is pointer only (licence: NONE) · get_code("b57d66d6dc47f45b")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from seketeam/deveval, 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.

count_indent Ran
ran, honoured the contract we drafted
seketeam/evocodebench/utils.py
code served (permissive licence) · get_code("e6083effdcd822ef")
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 os
import json
import tempfile
import textwrap

# Create a temporary directory and a test file with some indented code
temp_dir = tempfile.mkdtemp()

# Create a test code file with indented content
test_code = """def example_function():
    x = 1
    y = 2
    return x + y
"""

code_file_path = os.path.join(temp_dir, "test_code.py")
with open(code_file_path, 'w') as f:
    f.write(test_code)

# Create args object as a simple class
class Args:
    def __init__(self):
        self.source_code_root = temp_dir

# Create data dict
data = {
    'completion_path': 'test_code.py',
    'body_position': [2, 5]  # body starts at line 2 (1-indexed)
}

ARGS = {
    'args': Args(),
    'data': data
}
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 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
count_indent Ran
ran, honoured the contract we drafted
seketeam/evocodebench/prompt/process_completion.py
code served (permissive licence) · get_code("fb2bbc95a2bfbc79")
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 textwrap

def count_indent(code):
    if type(code) == str:
        return len(code) - len(textwrap.dedent(code))
    elif type(code) == list:
        for line in code:
            if line.strip() != '':
                return len(line) - len(textwrap.dedent(line))

# Test case 1: string with indentation
test_string = "    hello world"

# Test case 2: list of strings with indentation
test_list = ["    line 1", "    line 2", "no indent"]

ARGS = {
    "code": test_string
}
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 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
execution_tests Ran
ran, violated the contract we drafted
seketeam/evocodebench/pass_k.py
code served (permissive licence) · get_code("1e451f11c8932b9f")
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 subprocess
import psutil
import tempfile
import os

# Create a temporary directory to serve as project_path
temp_project = tempfile.mkdtemp()

# Create a minimal pytest test file
test_dir = os.path.join(temp_project, "tests")
os.makedirs(test_dir, exist_ok=True)

test_file_path = os.path.join(test_dir, "test_sample.py")
with open(test_file_path, "w") as f:
    f.write("""
def test_dummy():
    assert True
""")

# Create a minimal virtual environment structure (mock)
env_dir = os.path.join(temp_project, "myenv", "bin")
os.makedirs(env_dir, exist_ok=True)

# Create a dummy activate script
activate_script = os.path.join(env_dir, "activate")
with open(activate_script, "w") as f:
    f.write("#!/bin/bash\n")

# The test parameter should be a relative path to the pytest test
test = "tests/test_sample.py"

ARGS = {
    "test": test,
    "project_path": temp_project
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
func_timeout, psutil — 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
get_module_name Ran
ran; our contract draft was wrong, not the code
seketeam/evocodebench/parser/pyan_zyf_v2/anutils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from seketeam/deveval, is pointer only (licence: NONE) · get_code("aa151466ba7abc12")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from seketeam/deveval, 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.

head Ran
ran, violated the contract we drafted
seketeam/evocodebench/parser/pyan_zyf_v2/anutils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from seketeam/deveval, is pointer only (licence: NONE) · get_code("9b6c4ca7d09044ef")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from seketeam/deveval, 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.

load_api Ran
ran; our contract draft was wrong, not the code
seketeam/evocodebench/gpt_inference.py
code served (permissive licence) · get_code("4582b8be0cf47287")
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 os

# Create a temporary file with some API keys
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.txt')
temp_file.write("api_key_1\n")
temp_file.write("api_key_2\n")
temp_file.write("api_key_3\n")
temp_file.close()

ARGS = {
    'path': temp_file.name
}
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_file Ran
ran; our contract draft was wrong, not the code
seketeam/evocodebench/gpt_inference.py
code served (permissive licence) · get_code("d6407246571e12e2")
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 json
import os

# Create a temporary file with test data
temp_file = tempfile.NamedTemporaryFile(mode='w', delete=False, suffix='.jsonl')
temp_path = temp_file.name

# Write test data to the file
test_data = [
    {'namespace': 'namespace_1', 'other_field': 'value1'},
    {'namespace': 'namespace_2', 'other_field': 'value2'},
    {'namespace': 'namespace_3', 'other_field': 'value3'},
]

for item in test_data:
    temp_file.write(json.dumps(item) + '\n')

temp_file.close()

ARGS = {
    'path': temp_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
load_json_data Ran
ran; our contract draft was wrong, not the code
seketeam/evocodebench/utils.py
code served (permissive licence) · get_code("dbc6dc84387eb165")
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 os
import json
import tempfile

# Create a temporary JSON Lines file with sample data
temp_dir = tempfile.mkdtemp()
input_file = os.path.join(temp_dir, "test_data.jsonl")

# Write sample JSON Lines data
sample_data = [
    {"id": 1, "text": "hello world"},
    {"id": 2, "text": "foo bar"},
    {"id": 3, "text": "test data"}
]

with open(input_file, 'w') as f:
    for record in sample_data:
        f.write(json.dumps(record) + "\n")

ARGS = {
    "input_file": input_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
produce_prompt Ran
ran; our contract draft was wrong, not the code
seketeam/evocodebench/make_prompt.py
code served (permissive licence) · get_code("82e52f3a9b3bc119")
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 os
import tempfile

class MockTokenizer:
    def encode(self, text):
        # Simple mock: split by spaces and return token ids
        if not text:
            return []
        tokens = text.split()
        return list(range(len(tokens)))
    
    def decode(self, token_ids):
        # Simple mock: convert token ids back to text
        if not token_ids:
            return ""
        return " ".join([f"token_{i}" for i in token_ids])

class MockArgs:
    def __init__(self):
        self.context_window = 100
        self.max_tokens = 20
        self.setting = 'baseline'

# Create temporary prompt template files
temp_dir = tempfile.mkdtemp()
prompt_dir = os.path.join(temp_dir, 'prompt', 'template')

# Create baseline template
baseline_dir = os.path.join(prompt_dir, 'baseline')
os.makedirs(baseline_dir, exist_ok=True)
with open(os.path.join(baseline_dir, 'ChatLM.txt'), 'w') as f:
    f.write("Function: {function_name}\nCode:\n{input_code}")

# Create local_completion template
completion_dir = os.path.join(prompt_dir, 'local_completion')
os.makedirs(completion_dir, exist_ok=True)
with open(os.path.join(completion_dir, 'ChatLM.txt'), 'w') as f:
    f.write("Function: {function_name}\nContext Above:\n{contexts_above}\nCode:\n{input_code}")

# Create local_infilling template
infilling_dir = os.path.join(prompt_dir, 'local_infilling')
os.makedirs(infilling_dir, exist_ok=True)
with open(os.path.join(infilling_dir, 'ChatLM.txt'), 'w') as f:
    f.write("Function: {function_name}\nContext Above:\n{contexts_above}\nCode:\n{input_code}\nContext Below:\n{contexts_below}")

# Change to temp directory to find templates
os.chdir(temp_dir)

# Create mock data dict
d = {
    'input_code': 'def foo(): pass',
    'function_name': 'foo',
    'class_name': 'MyClass',
    'contexts_above': 'import sys\nimport os',
    'contexts_below': 'print("done")'
}

args = MockArgs()
tokenizer = MockTokenizer()

ARGS = {
    'args': args,
    'd': d,
    'tokenizer': tokenizer
}
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
retrieve_context_length Ran
ran, honoured the contract we drafted
seketeam/evocodebench/LM_inference.py
code served (permissive licence) · get_code("b3194a8fcb5677f4")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) 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.

ARGS = {
    'model_name': 'deepseek-coder-7b'
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
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 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
retrieve_special_ids Ran
ran, honoured the contract we drafted
seketeam/evocodebench/LM_inference.py
code served (permissive licence) · get_code("afdab5980a879713")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Sonnet 4.5, prompt python_fixture_v1) and returned tuple, ndim 1, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

class MockTokenizer:
    def __init__(self, model_name):
        self.model_name = model_name
        self.bos_token_id = 1
        
        # Set up token mappings based on model type
        if model_name.startswith('deepseek'):
            self.token_map = {
                '<|fim▁begin|>': 100,
                '<|fim▁hole|>': 101,
                '<|fim▁end|>': 102
            }
        elif model_name.startswith('codellama'):
            self.prefix_id = 200
            self.middle_id = 201
            self.suffix_id = 202
        elif model_name.startswith('starcoder2'):
            self.token_map = {
                '<fim_prefix>': 300,
                '<fim_middle>': 301,
                '<fim_suffix>': 302
            }
        else:
            self.token_map = {}
    
    def convert_tokens_to_ids(self, token):
        return self.token_map.get(token, -1)

# Test with a deepseek model
model_name = 'deepseek-coder-v2'
tokenizer = MockTokenizer(model_name)

ARGS = {
    'model_name': model_name,
    'tokenizer': tokenizer
}
Setup block drafted by
a language model, Claude Sonnet 4.5, prompt python_fixture_v1
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 int — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tail Ran
ran, violated the contract we drafted
seketeam/evocodebench/parser/pyan_zyf_v2/anutils.py
pointer only (licence: Apache-2.0); the stored copy get_code serves, from seketeam/deveval, is pointer only (licence: NONE) · get_code("0999d7eb180157c0")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (Apache-2.0), but get_code serves the stored copy, first harvested from seketeam/deveval, 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.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 8 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 8 later) and 7 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
adjust_indent 2 other papers
2 later
2405.19856 (2024-05)
DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories
2410.22821 (2024-10)
EvoCodeBench: An Evolving Code Generation Benchmark with Domain-Specific Evaluations
compute_pass_at_k 2 other papers
2 later
2405.19856 (2024-05)
DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories
2410.22821 (2024-10)
EvoCodeBench: An Evolving Code Generation Benchmark with Domain-Specific Evaluations
compute_recall 1 other paper
1 later
2405.19856 (2024-05)
DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories
get_module_name 1 other paper
1 later
2405.19856 (2024-05)
DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories
head 1 other paper
1 later
2405.19856 (2024-05)
DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories
retrieve_context_length 1 other paper
1 later
2410.22821 (2024-10)
EvoCodeBench: An Evolving Code Generation Benchmark with Domain-Specific Evaluations
retrieve_special_ids 1 other paper
1 later
2410.22821 (2024-10)
EvoCodeBench: An Evolving Code Generation Benchmark with Domain-Specific Evaluations
tail 1 other paper
1 later
2405.19856 (2024-05)
DevEval: A Manually-Annotated Code Generation Benchmark Aligned with Real-World Code Repositories

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 2026-11-21.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

How to evaluate Large Language Models (LLMs) in code generation is an open question. Existing benchmarks demonstrate poor alignment with real-world code repositories and are insufficient to evaluate the coding abilities of LLMs. This paper proposes a new benchmark - EvoCodeBench to address the preceding problems, which has three primary advances. (1) EvoCodeBench aligns with real-world repositories in multiple dimensions, e.g., code distributions and dependency distributions. (2) EvoCodeBench offers comprehensive annotations (e.g., requirements, reference code, and reference dependencies), and robust evaluation metrics (e.g., Pass@k and Recall@k). (3) EvoCodeBench is an evolving benchmark to avoid data leakage. We build an automatic pipeline to update EvoCodeBench from the latest repositories. We release the first version - EvoCodeBench-2403, containing 275 samples from 25 real-world repositories. Based on EvoCodeBench, we propose repository-level code generation and evaluate 10 popular LLMs (e.g., gpt-4, gpt-3.5, DeepSeek Coder, StarCoder 2, CodeLLaMa, Gemma, and Qwen 1.5). Our experiments reveal the coding abilities of these LLMs in real-world repositories. For example, the highest Pass@1 of gpt-4 only is 20.73% in our experiments. We also analyze failed cases and summarize the shortcomings of existing LLMs in EvoCodeBench. We release EvoCodeBench, all prompts, and LLMs' completions for further community analysis.

For agents

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

get_harvested_code_for_paper("2404.00599")
get_code_for_paper("2404.00599")
have("2404.00599")

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

curl https://syntology.ai/api/ran/2404.00599.json

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

[![Syntology run record](https://syntology.ai/api/ran/2404.00599.svg)](https://syntology.ai/paper/2404.00599)

Connect an agent — have() is free.