SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2506.20249 · 2025

Language Modeling by Language Models

arXiv · PDF · Open in the Atlas

Code that ran

5 of 6 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; 1 honoured the contract we drafted, 1 violated it and 0 have no contract verdict; 3 failures of our instrument; 1 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/2506.20249.json · badge: /api/ran/2506.20249.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
allenai/genesys canonical (classifier label)
classifier basis: census
5 ran · 1 no recorded run
FunctionStatusWhere it lives
dict_eq Ran
ran, violated the contract we drafted
allenai/genesys/model_discovery/utils.py
code served (permissive licence) · get_code("d6ac30a94d56959f")
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 torch
import numpy as np

torch.manual_seed(0)

# Analyze the function:
# dict_eq takes two dicts (dict1 and dict2) and compares them recursively
# It checks if keys exist, handles nested dicts, and compares values

# Create test dictionaries with various types of values
dict1 = {
    "key1": 42,
    "key2": "string_value",
    "key3": 3.14,
    "key4": [1, 2, 3],
    "nested": {
        "inner_key": "inner_value",
        "inner_num": 99
    }
}

dict2 = {
    "key1": 42,
    "key2": "string_value",
    "key3": 3.14,
    "key4": [1, 2, 3],
    "nested": {
        "inner_key": "inner_value",
        "inner_num": 99
    }
}

ARGS = {
    "dict1": dict1,
    "dict2": dict2
}
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 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
index_chunk_tool Ran
ran; our contract draft was wrong, not the code
allenai/genesys/model_discovery/evolution.py
code served (permissive licence) · get_code("df3caafe9e280578")
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
from unittest.mock import Mock, MagicMock

# Create mock Firestore document references and collections
class MockDocumentSnapshot:
    def __init__(self, data=None, exists_flag=True):
        self._data = data
        self._exists = exists_flag
    
    def exists(self):
        return self._exists
    
    @property
    def exists(self):
        return self._exists
    
    def to_dict(self):
        return self._data if self._data is not None else {}

class MockDocumentReference:
    def __init__(self, data=None, exists_flag=True):
        self._data = data if data is not None else {}
        self._exists = exists_flag
        self._set_data = {}
    
    def get(self):
        return MockDocumentSnapshot(self._data, self._exists)
    
    def set(self, data, merge=False):
        if merge:
            self._set_data.update(data)
            self._data.update(data)
        else:
            self._set_data = data
            self._data = data
        self._exists = True

class MockCollectionReference:
    def __init__(self):
        self._docs = {}
    
    def document(self, name):
        if name not in self._docs:
            self._docs[name] = MockDocumentReference(exists_flag=False)
        return self._docs[name]

# Set up the mock references
index_log_ref = MockDocumentReference(exists_flag=False)
index_collection_ref = MockCollectionReference()

# Pre-populate the index collection with an initial document
index_doc_ref = index_collection_ref.document('index')
index_doc_ref.set({'item_1': 'value_1', 'item_2': 'value_2'})

key = 'test_key'
chunk_size = 500

ARGS = {
    'index_log_ref': index_log_ref,
    'index_collection_ref': index_collection_ref,
    'key': key,
    'chunk_size': chunk_size
}
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, 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
letternum2num Ran
ran, honoured the contract we drafted
allenai/genesys/model_discovery/utils.py
code served (permissive licence) · get_code("49f874e390fe5027")
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 float, 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 torch
import numpy as np

torch.manual_seed(0)

# Analyze the function: letternum2num(s)
# Parameters:
#   - s: can be float, int, string with digits, or string ending with K/M/B/T
# The function converts letter-number strings to numeric values for sorting

ARGS = {
    "s": "1.5M"
}
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 float, 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
read_index Ran
ran; our contract draft was wrong, not the code
allenai/genesys/model_discovery/evolution.py
code served (permissive licence) · get_code("e4b2e38a8bb978ef")
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

# Mock Firestore document and collection classes
class MockDocument:
    def __init__(self, data):
        self._data = data
        self._exists = data is not None
    
    def exists(self):
        return self._exists
    
    def to_dict(self):
        return self._data if self._data is not None else {}

class MockDocumentRef:
    def __init__(self, doc_data):
        self._doc_data = doc_data
    
    def get(self):
        return MockDocument(self._doc_data)

class MockCollectionRef:
    def __init__(self, chunk_data):
        self._chunk_data = chunk_data
    
    def document(self, chunk_name):
        return MockDocumentRef(self._chunk_data.get(chunk_name))

# Set up mock data
# latest_index = 'index_1' means n_chunks = 2, so we need 'index' and 'index_1'
chunk_data = {
    'index': {'key1': 'value1', 'key2': 'value2'},
    'index_1': {'key3': 'value3', 'key4': 'value4'}
}

index_collection_ref = MockCollectionRef(chunk_data)
latest_index = 'index_1'

ARGS = {
    'latest_index': latest_index,
    'index_collection_ref': index_collection_ref
}
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
strscale Ran
ran; our contract draft was wrong, not the code
allenai/genesys/model_discovery/utils.py
code served (permissive licence) · get_code("1c79f1640254bc44")
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

ARGS = {
    'size': 5000000.0
}
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
theme_aware_options Not yet run
unverified (harvested, no recorded run)
allenai/genesys/bin/app_utils.py
code served (permissive licence) · get_code("8e03183878b9f4d8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 6 samples lifted from this paper's linked repositories, 6 carry a function identity and 0 do not. Of the 6 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 6 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 2026-11-21.

Repositories linked to this paper

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

Abstract

Can we leverage LLMs to model the process of discovering novel language model (LM) architectures? Inspired by real research, we propose a multi-agent LLM approach that simulates the conventional stages of research, from ideation and literature search (proposal stage) to design implementation (code generation), generative pre-training, and downstream evaluation (verification). Using ideas from scaling laws, our system, Genesys, employs a Ladder of Scales approach; new designs are proposed, adversarially reviewed, implemented, and selectively verified at increasingly larger model scales (14M$\sim$350M parameters) with a narrowing budget (the number of models we can train at each scale). To help make discovery efficient and factorizable, Genesys uses a novel genetic programming backbone, which we show has empirical advantages over commonly used direct prompt generation workflows (e.g., $\sim$86\% percentage point improvement in successful design generation, a key bottleneck). We report experiments involving 1,162 newly discovered designs (1,062 fully verified through pre-training) and find the best designs to be highly competitive with known architectures (e.g., outperform GPT2, Mamba2, etc., on 6/9 common benchmarks). We couple these results with comprehensive system-level ablations and formal results, which give broader insights into the design of effective autonomous discovery systems.

For agents

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

get_harvested_code_for_paper("2506.20249")
get_code_for_paper("2506.20249")
have("2506.20249")

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

curl https://syntology.ai/api/ran/2506.20249.json

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

[![Syntology run record](https://syntology.ai/api/ran/2506.20249.svg)](https://syntology.ai/paper/2506.20249)

Connect an agent — have() is free.