SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2501.18511 · ICML · 2025

WILDCHAT-50M: A Deep Dive Into the Role of Synthetic Data in Post-Training

Chinmay Hegde, Benjamin Feuer, M Meyer, K Chen, L Kaiser, M Wang, M Trebacz, G Zhao, G Leclerc, O Connell, K Lu, L Yang, and 2 more

arXiv · PDF · Open in the Atlas

Code that ran

6 of 9 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 6 have no contract verdict; 0 failures of our instrument; 3 no recorded run), measured 2026-11-22; 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/2501.18511.json · badge: /api/ran/2501.18511.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
penfever/wildchat-50m canonical (classifier label)
classifier basis not recorded on the link
6 ran · 3 no recorded run
FunctionStatusWhere it lives
convert_header_types Ran
ran on a synthesized input
penfever/wildchat-50m/src/core/utils.py
code served (permissive licence) · get_code("dd2d17638d3e997b")
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 Features. 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 datasets import Features, Value, Sequence

# Create a minimal features_dict to pass to convert_header_types
# The function accepts a features_dict parameter and returns a Features object
features_dict = {
    'conversation_hash': Value('string'),
    'timestamp': Value('timestamp[us, tz=UTC]'),
    'conversation': [{'content': Value('string')}],
    'turn': Value('int64'),
    'language': Value('string'),
    'toxic': Value('bool'),
    'redacted': Value('bool'),
    'state': Value('string'),
    'country': Value('string'),
    'hashed_ip': Value('string'),
    'header': {'accept-language': Value('string'), 'user-agent': Value('string')},
    'model': Value('string')
}

ARGS = {
    'features_dict': features_dict
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
datasets — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Features — a Features output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Features 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
get_features Ran
ran on a synthesized input
penfever/wildchat-50m/src/core/generate_model_responses.py
code served (permissive licence) · get_code("28053bf1dae0da2c")
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 Features. 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 datasets import Dataset, DatasetDict

# Create a simple Dataset with some features
data_dict = {
    "text": ["hello world", "foo bar"],
    "label": [0, 1]
}
dataset = Dataset.from_dict(data_dict)

# Test case 1: Pass a Dataset directly
ARGS = {
    "data": dataset
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
datasets — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Features — a Features output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Features 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
get_first_message_content Ran
ran on a synthesized input
penfever/wildchat-50m/src/util/length_control.py
code served (permissive licence) · get_code("ec7bca405fbfa692")
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
from typing import Optional, Set, List, Dict

torch.manual_seed(0)

# Construct a valid conversation list with message dictionaries
conversation = [
    {
        'content': 'Hello, this is the first message',
        'role': 'user',
        'timestamp': 1234567890
    },
    {
        'content': 'This is the second message',
        'role': 'assistant',
        'timestamp': 1234567891
    },
    {
        'content': 'This is the third message',
        'role': 'user',
        'timestamp': 1234567892
    }
]

ARGS = {
    'conversation': conversation
}
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
process_list_dicts_column Ran
ran on a synthesized input
penfever/wildchat-50m/src/util/remove_cols_from_hf_dataset.py
code served (permissive licence) · get_code("6c853e15bbc928ff")
How it ran

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

import torch
import numpy as np

conv_keys = ['content', 'finish_reason', 'hashed_ip', 'role', 'toxic']

# Create a minimal example dataset entry
example = {
    'conversation': [
        {
            'content': 'Hello, how are you?',
            'finish_reason': 'stop',
            'hashed_ip': 'abc123',
            'role': 'user',
            'toxic': False
        },
        {
            'content': 'I am doing well, thank you!',
            'finish_reason': 'stop',
            'hashed_ip': 'def456',
            'role': 'assistant',
            'toxic': False
        }
    ]
}

ARGS = {
    'example': example,
    'column_name': 'conversation',
    'target_column': 'conversation'
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
push_to_hub_fallback Ran
ran on a synthesized input
penfever/wildchat-50m/src/core/generate_model_responses.py
code served (permissive licence) · get_code("4c25d68b54111838")
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 time
from unittest.mock import Mock, MagicMock
from huggingface_hub.errors import HfHubHTTPError

# Create a mock dataset with push_to_hub method
mock_dataset = Mock()
mock_dataset.push_to_hub = MagicMock(return_value=None)

# Create arguments dict
ARGS = {
    "response_dataset": mock_dataset,
    "repo_id": "user/test-repo",
    "max_retries": 3,
    "delay": 60
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
huggingface_hub — 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
update_nested_field_batch Ran
ran on a synthesized input
penfever/wildchat-50m/src/core/generate_model_responses.py
code served (permissive licence) · get_code("2718a0c2fcafd8fd")
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 Dataset. 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 datasets import Dataset
from unittest.mock import MagicMock

torch_manual_seed = 0

# Create a minimal args object with all necessary attributes
class Args:
    def __init__(self):
        self.gen_judgments = False
        self.model_name = "test_model"
        self.reference_column = "conversation"
        self.judgment_factor = "factor_1"

args = Args()

# Create sample conversation data
sample_conversations = [
    [
        {"content_token_ids": [1, 2, 3], "role": "user", "content": "Hello"},
        {"content_token_ids": [4, 5], "role": "assistant", "content": "Hi there"}
    ],
    [
        {"content_token_ids": [10, 11], "role": "user", "content": "How are you?"},
        {"content_token_ids": [12, 13, 14], "role": "assistant", "content": "I'm fine"}
    ]
]

# Create a Dataset with conversation data
dataset = Dataset.from_dict({
    "conversation": sample_conversations,
    "id": [0, 1]
})

# Create new_values dict mapping token_id tuples to replacement content
new_values = {
    (1, 2, 3): ["Updated response 1", 0.95, "stop"],
    (10, 11): ["Updated response 2", 0.87, "length"]
}

# Default value for logprob
default_value_for_logprob = "-1.0"

ARGS = {
    "dataset": dataset,
    "new_values": new_values,
    "default_value_for_logprob": default_value_for_logprob,
    "args": args
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
datasets — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Dataset — a Dataset output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Dataset output the recorded ndim 1 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
convert_dataset_types Not yet run
unverified (harvested, no recorded run)
penfever/wildchat-50m/src/core/utils.py
code served (permissive licence) · get_code("85c0af3e2be31833")
How it ran

no invocation recorded.

get_judge_template Not yet run
unverified (harvested, no recorded run)
penfever/wildchat-50m/src/core/judgment.py
code served (permissive licence) · get_code("19bfb0450edccbaa")
How it ran

no invocation recorded.

is_unique Not yet run
unverified (harvested, no recorded run)
penfever/wildchat-50m/src/util/length_control.py
code served (permissive licence) · get_code("3f6a8cba60dab66b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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 9 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

Language model (LLM) post-training, from DPO to distillation, can refine behaviors and unlock new skills, but the open science supporting these post-training techniques is still in its infancy. One limiting factor has been the difficulty of conducting large-scale comparative analyses of synthetic data generating models and LLM judges. To close this gap, we introduce WILDCHAT-50M, the largest public chat dataset to date. We extend the existing WildChat dataset to include responses not only from GPT, but from over 50 different open-weight models, ranging in size from 0.5B to 104B parameters. We conduct an extensive comparative analysis and demonstrate the potential of this dataset by creating RE-WILD, our own public SFT mix, which outperforms the recent Tulu-3 SFT mixture from Allen AI with only 40% as many samples. Our dataset, samples and code are available at https://github. com/penfever/wildchat-50m.

For agents

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

get_harvested_code_for_paper("2501.18511")
get_code_for_paper("2501.18511")
have("2501.18511")

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

curl https://syntology.ai/api/ran/2501.18511.json

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

[![Syntology run record](https://syntology.ai/api/ran/2501.18511.svg)](https://syntology.ai/paper/2501.18511)

Connect an agent — have() is free.