SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.01006 · NeurIPS · 2024

SEMCODER: Training Code Language Models with Comprehensive Semantics Reasoning

Junfeng Yang, Yangruibo Ding, Marcus Min, Gail Kaiser, Baishakhi Ray, Jinjun Peng

arXiv · PDF · Open in the Atlas

Code that ran

12 of 15 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 10 failures of our instrument; 3 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2406.01006.json · badge: /api/ran/2406.01006.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
ARiSE-Lab/SemCoder canonical (classifier label)
classifier basis: census
9 ran · 2 no recorded run
arise-lab/semcoder community
Papers with Code archive
3 ran · 1 no recorded run
FunctionStatusWhere it lives
attach_execution_exception Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/experiments/build_self_refine_input.py
code served (permissive licence) · get_code("23ef3078d45c966b")
How it ran

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

import re

# Construct the arguments for attach_execution_exception

# traced_program should be a multi-line string containing Python code
traced_program = """def test_function():
    x = 1 + 1
    y = x / 0
    z = y + 2
    return z"""

# trace should be a multi-line string containing execution trace information
# with lines that may contain "exception" keywords and exception messages
trace = """File "test.py", line 1
  exception 3
Exception: division by zero
File "test.py", line 4
  exception 5
Exception: type error"""

ARGS = {
    "traced_program": traced_program,
    "trace": trace,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
chunked Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/src/utils.py
code served (permissive licence) · get_code("8fbbf1d9e53be9ec")
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 generator. 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 chunks a sequence into n-sized chunks
# It requires:
# - seq: a Sequence (list, tuple, string, etc.)
# - n: an int representing chunk size

# Let's create a simple list as the sequence
seq = list(range(10))  # [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
n = 3  # chunk size of 3

ARGS = {
    "seq": seq,
    "n": n,
}
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 generator — a generator output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a generator 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
complete_code Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/experiments/cruxeval_utils.py
code served (permissive licence) · get_code("b9090aa3af341871")
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
from collections import defaultdict
import tempfile
import os

torch.manual_seed(0)

# Stand-in task object with postprocess_generation method
class Task:
    def postprocess_generation(self, text, task_idx):
        return text.strip()

# Stand-in model with generate method
class Model(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def generate(self, prompt_token_ids, sampling_params, use_tqdm=False):
        # Returns a list containing a single object with outputs attribute
        class Output:
            def __init__(self, text):
                self.text = text
        
        class OutputGroup:
            def __init__(self, outputs):
                self.outputs = outputs
        
        # Generate 2 outputs (matching batch_size)
        outputs = [Output(f" generated text {i}") for i in range(2)]
        return [OutputGroup(outputs)]

# Stand-in sampling params
class SamplingParams:
    def __init__(self):
        self.max_tokens = 20

# Stand-in dataset
class Dataset:
    def __init__(self):
        self.n_copies = 1
    
    def __len__(self):
        return 2

# Stand-in dataloader that yields batches
class Dataloader:
    def __init__(self):
        self.dataset = Dataset()
    
    def __iter__(self):
        # Yield 2 batches
        for batch_idx in range(2):
            batch = {
                'ids': torch.randint(0, 100, (2, 10)),  # batch_size=2, max_seq_len=10
                'input_len': 8,
                'row_index': torch.tensor([batch_idx, batch_idx]),
                'prompt': [f"prompt {batch_idx}"]
            }
            yield batch

# Create temporary log file
temp_dir = tempfile.mkdtemp()
log_path = os.path.join(temp_dir, "test_log.jsonl")

# Construct ARGS
ARGS = {
    'task': Task(),
    'model': Model(),
    'sampling_params': SamplingParams(),
    'dataloader': Dataloader(),
    'batch_size': 2,
    'n_tasks': 2,
    'log_path': log_path,
    'prefix': 'test_',
    'postprocess': True
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — 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
concise_spad Ran
ran; our contract draft was wrong, not the code
arise-lab/semcoder/experiments/construct_trace.py
code served (permissive licence) · get_code("7e638715779072af")
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 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

# The function takes a single parameter: line (a string)
# It processes the string to find [STATE markers and replace intermediate ones with '...'
# We need to create a test string with multiple [STATE markers

# Create a line with more than 3 [STATE markers
line = "[STATE 0] data [STATE 1] more [STATE 2] content [STATE 3] extra [STATE 4] final [STATE 5] end"

ARGS = {
    "line": line
}
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 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
convertv2s Ran
ran; our contract draft was wrong, not the code
arise-lab/semcoder/experiments/construct_trace.py
code served (permissive licence) · get_code("5c7a36496bbe50b3")
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 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 json

torch.manual_seed(0)

ARGS = {
    "obj": {
        "key1": "value1",
        "key2": 42,
        "key3": [1, 2, 3],
        "key4": True,
        "key5": None
    },
    "output": False
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_failed_test Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/experiments/build_self_refine_input.py
code served (permissive licence) · get_code("660d74f09a475daa")
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

def extract_failed_test(traced_program, funcname):
    """
    extract the failed test case from the traced program
    """
    failed_test = ''
    for line in traced_program.split('\n'):
        if line.lstrip().startswith(f'retv = {funcname}'):
            if line.split(',')[-1] == ' )':
                line = ','.join(line.split(',')[:-1]) + ')'
            failed_test += line.lstrip() + '\n'
        if line.lstrip().startswith('assert retv == '):
            failed_test += line.lstrip() + '\n'
    return failed_test


# Construct valid arguments
traced_program = """some_code_here()
    retv = my_function(arg1, arg2, arg3 )
    result = process(retv)
    assert retv == expected_value
    more_code()
"""

funcname = "my_function"

ARGS = {
    "traced_program": traced_program,
    "funcname": funcname
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
extract_trace Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/experiments/build_self_refine_input.py
code served (permissive licence) · get_code("7e07a365b5619a81")
How it ran

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

import re

def extract_trace(trace, inference=False):
    """
    extract the trace of the real function, remove the other utility information
    """
    trace_lines = trace.split('\n')
    input_pattern = re.compile('Starting var:\\.* (.*?) = (.*)')
    lines_w_indent = []
    for idx, line in enumerate(trace_lines):
        if line.startswith('    '):
            lines_w_indent.append(idx)
            if (input_match := input_pattern.match(line.lstrip())):
                input_info = input_match.group(2)
                if '...' in input_info and (not inference):
                    return None
                if len(input_info) > 100 and (not inference):
                    return None
    if len(lines_w_indent) == 0:
        if not inference:
            return None
        else:
            return trace
    end = -1
    for i in range(len(lines_w_indent) - 1):
        if lines_w_indent[i] + 1 != lines_w_indent[i + 1]:
            if not inference:
                return None
            else:
                end = lines_w_indent[i]
    start = lines_w_indent[0]
    if end == -1:
        end = lines_w_indent[-1]
    trace = '\n'.join([l.lstrip() for l in trace_lines[start:end + 1]])
    return trace


trace_simple = """Some header info
    Starting var: x = tensor([1, 2, 3])
    y = x + 1
    z = y * 2
Some footer info"""

trace_with_long_input = """Some header
    Starting var: very_long_var = this_is_a_very_long_string_that_exceeds_one_hundred_characters_in_length_to_test_the_filtering_condition_for_long_inputs_in_the_function
Some footer"""

trace_with_ellipsis = """Header
    Starting var: data = [1, 2, ..., 100]
Footer"""

trace_inference = """Preamble
    Starting var: result = some_data_here
    line2 = process(result)
More stuff"""

ARGS = {
    "trace": trace_simple,
    "inference": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
handle_value Ran
ran, honoured the contract we drafted
arise-lab/semcoder/experiments/construct_trace.py
code served (permissive licence) · get_code("6834377c553301df")
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 list, ndim 1, dtype family int; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import json

def handle_value(var_value):
    """
    Handle value as json object in the trace log.
    """
    if var_value.startswith('[') and var_value.endswith(']'):
        var_value = var_value.replace("'", '"')
    try:
        var_value = json.loads(var_value)
        return var_value
    except:
        if var_value == 'None':
            return None
        elif var_value == "''":
            return ''
        elif var_value == '[]':
            return []
        elif var_value == '{}':
            return {}
        elif var_value == 'True':
            return True
        elif var_value == 'False':
            return False
        elif var_value.startswith("'"):
            return var_value[1:-1]
        else:
            return var_value


ARGS = {
    "var_value": "[1, 2, 3]"
}
Setup block drafted by
a language model, Claude Haiku 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 list, ndim 1, dtype family int — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_cot_output_prompt Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/experiments/cruxeval_prompts.py
code served (permissive licence) · get_code("b4a6d7a477e69976")
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

# Define the arguments for make_cot_output_prompt
# The function takes a single parameter 's' which is a tuple of (code, input)

code = """def f(x):
    return x * 2 + 1"""

input_str = "5"

s = (code, input_str)

ARGS = {
    "s": s
}
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
make_direct_output_prompt Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/experiments/cruxeval_prompts.py
code served (permissive licence) · get_code("7ff4e308c6863412")
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

# The function make_direct_output_prompt takes a single parameter 's'
# which is a tuple of (code, input)
# where code is a string containing Python code
# and input is a string representing the input to the function

code_example = """def f(n):
    return n * 2"""

input_example = "5"

s = (code_example, input_example)

ARGS = {
    "s": s
}
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
make_forward_monologue_output_prompt Ran
ran; our contract draft was wrong, not the code
ARiSE-Lab/SemCoder/experiments/cruxeval_prompts.py
code served (permissive licence) · get_code("1c29bebc3b2830c3")
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

# Analyzing the function: make_forward_monologue_output_prompt(s)
# The parameter 's' is unpacked as: code, input = s
# So 's' should be a tuple/list of (code, input)
# - code: a string containing Python code (with newlines)
# - input: a string representing function input

# Creating a simple Python code string
code_str = """def add(x, y):
    return x + y"""

# Creating a simple input string
input_str = "1, 2"

# Constructing the tuple
s = (code_str, input_str)

ARGS = {
    's': s
}
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
pad_sequences Ran
ran, honoured the contract we drafted
ARiSE-Lab/SemCoder/src/wrapper.py
code served (permissive licence) · get_code("7b178c461482e3f6")
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 (3, 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)

ARGS = {
    "sequences": [[1, 2, 3], [4, 5], [6, 7, 8, 9]],
    "pad_value": 0,
    "padding_side": "right",
    "dtype": torch.long,
    "padding_length": None,
}
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 (3, 4) — shape (3, 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
attach_trace_to_code Not yet run
unverified (harvested, no recorded run)
arise-lab/semcoder/experiments/construct_trace.py
code served (permissive licence) · get_code("926bd7fef63dd64e")
How it ran

no invocation recorded.

get_model_context Not yet run
unverified (harvested, no recorded run)
ARiSE-Lab/SemCoder/src/wrapper.py
code served (permissive licence) · get_code("da99ae3e6a8eb89e")
How it ran

no invocation recorded.

get_task Not yet run
unverified (harvested, no recorded run)
ARiSE-Lab/SemCoder/experiments/cruxeval_utils.py
code served (permissive licence) · get_code("0ea1d7ce8d9d3e2a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 15 samples lifted from this paper's linked repositories, 15 carry a function identity and 0 do not. Of the 15 distinct functions with an identity, 2 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 2 later) and 13 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
chunked 3 other papers
3 later
2410.24198 (2024-10)
SelfCodeAlign: Self-Alignment for Code Generation
2412.21199 (2024-12)
HumanEval Pro and MBPP Pro: Evaluating Large Language Models on Self-invoking Code Generation
2609.12584 (2026-09)
Clustering-Based Balanced Sampling and Allocation with Data Parallelism for High-Performance Fine-Tuning
pad_sequences 2 other papers
2 later
2410.24198 (2024-10)
SelfCodeAlign: Self-Alignment for Code Generation
2609.12584 (2026-09)
Clustering-Based Balanced Sampling and Allocation with Data Parallelism for High-Performance Fine-Tuning

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

Code Large Language Models (Code LLMs) have excelled at tasks like code completion but often miss deeper semantics such as execution effects and dynamic states. This paper aims to bridge the gap between Code LLMs' reliance on static text data and the need for semantic understanding for complex tasks like debugging and program repair. We introduce a novel strategy, monologue reasoning, to train Code LLMs to reason comprehensive semantics, encompassing high-level functional descriptions, local execution effects of individual statements, and overall input/output behavior, thereby linking static code text with dynamic execution states. We begin by collecting PYX, a clean Python corpus of fully executable code samples with functional descriptions and test cases. We propose training Code LLMs not only to write code but also to understand code semantics by reasoning about key properties, constraints, and execution behaviors using natural language, mimicking human verbal debugging, i.e., rubber-duck debugging. This approach led to the development of SEMCODER, a Code LLM with only 6.7B parameters, which shows competitive performance with GPT-3.5-turbo on code generation and execution reasoning tasks. SEMCODER achieves 79.3% on HumanEval (GPT-3.5-turbo: 76.8%), 63.6% on CRUXEval-I (GPT-3.5-turbo: 50.3%), and 63.9% on CRUXEval-O (GPT-3.5turbo: 59.0%). We also study the effectiveness of SEMCODER's monologue-style execution reasoning compared to concrete scratchpad reasoning, showing that our approach integrates semantics from multiple dimensions more smoothly. Finally, we demonstrate the potential of applying learned semantics to improve Code LLMs' debugging and self-refining capabilities. Our data, code, and models are available at: https://github.com/ARiSE-Lab/SemCoder. We show that, by training on this approach, SEMCODER can generate, reason about execution, debug and refine code in a more intuitive and effective manner, pushing the boundaries of what current Code LMs can achieve in different software engineering tasks. Performance of SEMCODER SEMCODER, while having only 6.7B parameters, exhibits exceptional performance in code generation and execution reasoning tasks, surpassing larger models like GPT-3.5-turbo and various open-source models. For code generation, SEMCODER variants achieve a pass@1 of 79.3% on HumanEval [1

For agents

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

get_harvested_code_for_paper("2406.01006")
get_code_for_paper("2406.01006")
have("2406.01006")

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

curl https://syntology.ai/api/ran/2406.01006.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.01006.svg)](https://syntology.ai/paper/2406.01006)

Connect an agent — have() is free.