SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2006.16205 · ICML · 2021

Composed Fine-Tuning: Freezing Pre-Trained Denoising Autoencoders for Improved Generalization

Sang Xie, Percy Liang, Tengyu Ma

arXiv · PDF · Open in the Atlas

Code that ran

14 of 14 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 12 have no contract verdict; 2 failures of our instrument; 0 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/2006.16205.json · badge: /api/ran/2006.16205.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
p-lambda/unlabeled_outputs canonical (classifier label)
classifier basis: census
12 ran · 0 no recorded run
p-lambda/composed_finetuning canonical (classifier label)
classifier basis: classifier_2025
2 ran · 0 no recorded run
FunctionStatusWhere it lives
add_newlines Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/pretty_print.py
code served (permissive licence) · get_code("e22284ad0b102063")
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 add_newlines(string):
    new_string = ''
    for_started = False
    if_started = False
    num_open_parens = 0
    new_string = string.replace('$', '\n')
    new_string = new_string.replace('~', '\t')
    return new_string

ARGS = {
    'string': 'hello$world~test$case~foo'
}
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
concat_code Ran
ran; our contract draft was wrong, not the code
p-lambda/composed_finetuning/code/evaluate_full.py
code served (permissive licence) · get_code("c60712bd3a0d1e33")
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 pandas as pd

# Create a small DataFrame with a 'code' column
# The function iterates through rows and concatenates the 'code' column values
data = {
    'code': ['print("hello")', 'x = 5', 'y = x + 1']
}
curr_df = pd.DataFrame(data)

ARGS = {
    'curr_df': curr_df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — 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
concat_code Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/run_spoc.py
code served (permissive licence) · get_code("49d63b429e4a4074")
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 pandas as pd
import torch
import numpy as np

torch.manual_seed(0)

class MockTokenizer:
    """Stand-in tokenizer that implements tokenize method."""
    def tokenize(self, text):
        # Return a list containing a list of tokens
        # The function expects tokenizer.tokenize(...)[0] to work
        tokens = text.split()
        return [tokens]

class MockRow:
    """Stand-in for a DataFrame row with code and indent attributes."""
    def __init__(self, code, indent):
        self.code = code
        self.indent = indent

# Create a mock DataFrame with the required structure
curr_df = pd.DataFrame({
    'code': ['def hello world', 'return x y z', 'print test'],
    'indent': [0, 1, 2]
})

# Add code and indent as attributes that iloc will access correctly
for i, row in curr_df.iterrows():
    curr_df.at[i, 'code'] = ['def hello world', 'return x y z', 'print test'][i]
    curr_df.at[i, 'indent'] = [0, 1, 2][i]

# Create tokenizer
tokenizer = MockTokenizer()

ARGS = {
    'curr_df': curr_df,
    'tokenizer': tokenizer
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, pandas, 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
concat_text Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/run_spoc.py
code served (permissive licence) · get_code("459ab85d861ca156")
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 pandas as pd

# Create a small DataFrame with 'text' and 'indent' columns
curr_df = pd.DataFrame({
    'text': ['Hello world', 'Second line', float('nan'), 'Third line'],
    'indent': [0, 1, 2, 3]
})

ARGS = {
    'curr_df': curr_df
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — 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
encode Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/run_synthetic.py
code served (permissive licence) · get_code("465db07d479dc049")
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.

from pathlib import Path
import tempfile

# Create temporary directory for test files
temp_dir = tempfile.mkdtemp()

# Create dummy input files
input_file_1 = Path(temp_dir) / "input1.txt"
input_file_2 = Path(temp_dir) / "input2.txt"
input_file_1.write_text("sample text for encoding")
input_file_2.write_text("another sample text")

# Create dummy SPM model files (just empty files for the purpose of this setup)
spm_model_prefix = str(Path(temp_dir) / "test_model")
Path(f"{spm_model_prefix}.model").write_text("dummy model")

# Define input paths as Path objects
in_paths = [input_file_1, input_file_2]

ARGS = {
    "in_paths": in_paths,
    "spm_model_prefix": spm_model_prefix
}
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
fix_char_string_tok Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/format_pairs.py
code served (permissive licence) · get_code("a0d03dc853adfbd0")
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.

# Setup code for fix_char_string_tok function

def fix_char_string_tok(tokens):
    res_tokens = []
    if tokens and tokens[0] == '}':
        tokens = tokens[1:]
    if tokens and tokens[-1] == '{':
        tokens = tokens[:-1]
    for token in tokens:
        if token[0] == '"' and token[-1] == '"':
            res_tokens.append('"')
            res_tokens.append(token[1:-1])
            res_tokens.append('"')
        elif token[0] == "'" and token[-1] == "'":
            res_tokens.append("'")
            res_tokens.append(token[1:-1])
            res_tokens.append("'")
        else:
            res_tokens.append(token)
    return res_tokens


# Build test arguments
ARGS = {
    "tokens": [
        '"hello"',
        "'world'",
        'plain_token',
        '"quoted_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 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
get_lens Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/fairseq_modules/criterion.py
code served (permissive licence) · get_code("cd0dc7e8ddbce005")
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 1, dtype family int, shape (2,) 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)

# Create a small batch of token sequences
# Shape: (batch_size=2, sequence_length=4)
tokens = torch.tensor([
    [1, 2, 3, 0],
    [5, 6, 0, 0]
], dtype=torch.long)

# Padding index
pad_idx = 0

ARGS = {
    "tokens": tokens,
    "pad_idx": pad_idx
}
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 1, dtype family int, shape (2,) — shape (2,) 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
pad Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/fairseq_modules/criterion.py
code served (permissive licence) · get_code("245dba3efa0406c1")
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. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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)

# Create a small batch of token sequences
# batch_size=2, sequence_length=4
tokens = torch.tensor([
    [10, 20, 2, 0],  # sequence with eos_idx=2 at position 2
    [10, 20, 30, 2]  # sequence with eos_idx=2 at position 3
])

pad_idx = 0
eos_idx = 2

ARGS = {
    "tokens": tokens,
    "pad_idx": pad_idx,
    "eos_idx": eos_idx
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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_before Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/fairseq_modules/criterion.py
code served (permissive licence) · get_code("b9c973b50d3ba0a3")
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 (2, 5) as a later run recorded it. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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
import torch

torch.manual_seed(0)

# Create sample token lists as tensors
# Two sequences with different lengths, containing token indices
tokens = [
    torch.tensor([10, 11, 12, 13], dtype=torch.long),
    torch.tensor([20, 21], dtype=torch.long),
]

pad_idx = 0
eos_idx = 99

ARGS = {
    "tokens": tokens,
    "pad_idx": pad_idx,
    "eos_idx": eos_idx,
    "return_orig_mask": False,
    "add_eos": True,
}
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 Tensor, ndim 2, dtype family int, shape (2, 5) — shape (2, 5) 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
parse_pred_file Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/run_synthetic.py
code served (permissive licence) · get_code("8f8d509260110fb3")
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 PosixPath. 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.

from pathlib import Path
import tempfile
import os

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

# Create a sample prediction file with the expected format
pred_file_path = Path(temp_dir) / "predictions.txt"
with open(pred_file_path, 'w') as f:
    f.write("H-0\t0.95\tpredicted output 1\n")
    f.write("some other line\n")
    f.write("H-1\t0.87\tpredicted output 2\n")
    f.write("H-2\t0.92\tanother prediction\n")

ARGS = {
    'pred_file': pred_file_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 PosixPath — a PosixPath output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PosixPath 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_preds Ran
ran; our contract draft was wrong, not the code
p-lambda/composed_finetuning/code/evaluate_full.py
code served (permissive licence) · get_code("60931bcb2604c46d")
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
from pathlib import Path

# Create a temporary file with predictions in the expected format
temp_dir = tempfile.mkdtemp()
temp_file = Path(temp_dir) / "predictions.txt"

# Write test data in the format expected by parse_preds
# Format: H-<idx>\t<prob>\t<prediction>
with open(temp_file, 'w') as f:
    f.write("H-0\t0.95\tprediction one\n")
    f.write("H-2\t0.87\tprediction three\n")
    f.write("H-1\t0.92\tprediction two\n")
    f.write("some other line\n")
    f.write("H-3\t0.88\tprediction four\n")

ARGS = {
    'path': str(temp_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
strip_ext Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/run_spoc.py
code served (permissive licence) · get_code("18e78e1f6ef94f00")
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 PosixPath. 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.

from pathlib import Path

ARGS = {
    "path": Path("/home/user/documents/file.txt")
}
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 PosixPath — a PosixPath output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PosixPath output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
tokenize_code Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/format_pairs.py
code served (permissive licence) · get_code("f4e1ee0459a1e883")
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 torch
import numpy as np

# The function tokenize_code takes C++ code as a string and tokenizes it
# using clang.cindex. However, since we don't have clang available in a sandbox,
# we'll create a minimal valid input that demonstrates the function's structure.

# The function expects:
# - code: a string containing C++ code

# Let's create a simple C++ code string
simple_cpp_code = '''
int main() {
    const char* str = "hello world";
    return 0;
}
'''

ARGS = {
    "code": simple_cpp_code
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
clang, numpy, torch — 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
tokenize_text Ran
ran on a synthesized input
p-lambda/unlabeled_outputs/code/format_pairs.py
code served (permissive licence) · get_code("6708dda8d2ddd583")
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 re

TEXT_TOKENIZER = re.compile('\\w+|[^\\w\\s]', re.UNICODE)

def tokenize_text(text):
    return TEXT_TOKENIZER.findall(text)

ARGS = {
    'text': 'Hello, world! This is a test.'
}
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

Also present in other papers' code

Of 14 samples lifted from this paper's linked repositories, 14 carry a function identity and 0 do not. Of the 14 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 14 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.

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

We focus on prediction problems with structured outputs that are subject to output validity constraints, e.g. pseudocode-to-code translation where the code must compile. While labeled input-output pairs are expensive to obtain, "unlabeled" outputs, i.e. outputs without corresponding inputs, are freely available (e.g. code on GitHub) and provide information about output validity. We can capture the output structure by pre-training a denoiser to denoise corrupted versions of unlabeled outputs. We first show that standard fine-tuning after pre-training destroys some of this structure. We then propose composed fine-tuning, which fine-tunes a predictor composed with the pre-trained denoiser, which is frozen to preserve output structure. For two-layer ReLU networks, we prove that composed fine-tuning significantly reduces the complexity of the predictor, thus improving generalization. Empirically, we show that composed fine-tuning improves over standard fine-tuning on two pseudocode-to-code translation datasets (3% and 6% relative). The improvement from composed fine-tuning is magnified on out-of-distribution (OOD) examples (4% and 25% relative).

For agents

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

get_harvested_code_for_paper("2006.16205")
get_code_for_paper("2006.16205")
have("2006.16205")

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

curl https://syntology.ai/api/ran/2006.16205.json

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

[![Syntology run record](https://syntology.ai/api/ran/2006.16205.svg)](https://syntology.ai/paper/2006.16205)

Connect an agent — have() is free.