SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2402.12030 · 2024

Towards Cross-Tokenizer Distillation: the Universal Logit Distillation Loss for LLMs

arXiv · PDF · Open in the Atlas

Code that ran

14 of 19 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 7 have no contract verdict; 5 failures of our instrument; 5 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/2402.12030.json · badge: /api/ran/2402.12030.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
nicolas-bzrd/llm-recipes canonical (classifier label)
classifier basis: census
6 ran · 2 no recorded run
arcee-ai/distillkit community (classifier label)
classifier basis not recorded on the link
8 ran · 3 no recorded run
FunctionStatusWhere it lives
byte2gb Ran
ran, honoured the contract we drafted
nicolas-bzrd/llm-recipes/models/memory.py
pointer only (licence: NONE) · get_code("6ef625d775286fe7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

byte2mb Ran
ran, honoured the contract we drafted
nicolas-bzrd/llm-recipes/models/memory.py
pointer only (licence: NONE) · get_code("e20e1a734f7d63bc")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

dense_js_div Ran
ran on a synthesized input
arcee-ai/distillkit/distillkit/lossfuncs/jsd.py
code served (permissive licence) · get_code("2e53e5a9faf149a2")
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 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 math

torch.manual_seed(0)

# Create small tensors for logits
batch_size = 2
seq_len = 3
vocab_size = 4

logits = torch.randn(batch_size, seq_len, vocab_size)
target_logits = torch.randn(batch_size, seq_len, vocab_size)

# Create a mask tensor (True for tokens to include)
mask = torch.ones(batch_size, seq_len, dtype=torch.bool)

temperature = 1.0

ARGS = {
    "logits": logits,
    "target_logits": target_logits,
    "mask": mask,
    "temperature": temperature,
}
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 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
dense_tvd Ran
ran on a synthesized input
arcee-ai/distillkit/distillkit/lossfuncs/tvd.py
code served (permissive licence) · get_code("5713ef284636287e")
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 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

torch.manual_seed(0)

# Small tensors with distinct dimensions
batch_size = 2
seq_len = 3
vocab_size = 4

logits = torch.randn(batch_size, seq_len, vocab_size)
target_logits = torch.randn(batch_size, seq_len, vocab_size)
mask = torch.ones(batch_size, seq_len)
temperature = 1.0

ARGS = {
    "logits": logits,
    "target_logits": target_logits,
    "mask": mask,
    "temperature": temperature,
}
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 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
fsdp_auto_wrap_policy Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
nicolas-bzrd/llm-recipes/models/fsdp.py
pointer only (licence: NONE) · get_code("ea0600a9b9ad4670")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_max_seqlen_in_batch Ran
ran; our contract draft was wrong, not the code
arcee-ai/distillkit/distillkit/monkey_patch_packing.py
code served (permissive licence) · get_code("d4a53ff33a6fcd12")
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 (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)

# The function expects an attention_mask tensor
# It finds max value, then counts occurrences of each value from 1 to max
# So we need a tensor with integer values where max value determines the loop count

# Create a small attention mask tensor with shape (batch_size, seq_len)
# Values should be integers from 0 to some max value
attention_mask = torch.tensor([
    [1, 2, 2, 0],
    [1, 1, 2, 2]
], dtype=torch.int64)

ARGS = {
    "attention_mask": attention_mask
}
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 (4,) — shape (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
get_parameter_dtypes Ran
ran; our contract draft was wrong, not the code
nicolas-bzrd/llm-recipes/models/tools.py
pointer only (licence: NONE) · get_code("fea76732cc6f16f5")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

get_unpad_data Ran
ran on a synthesized input
arcee-ai/distillkit/distillkit/monkey_patch_packing.py
code served (permissive licence) · get_code("0616e148972493f0")
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. 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 torch.nn.functional as F

torch.manual_seed(0)

# Create a small attention_mask tensor
# The function expects an attention_mask where values indicate sequence positions
# Values range from 0 to some max_num, where non-zero values indicate valid tokens
# Shape: (batch_size, seq_len) = (2, 4)
attention_mask = torch.tensor([
    [1, 2, 3, 0],
    [1, 2, 0, 0]
], dtype=torch.int32)

ARGS = {
    "attention_mask": attention_mask
}
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
load_balancing_loss_func Ran
ran on a synthesized input
arcee-ai/distillkit/distillkit/monkey_patch_packing.py
code served (permissive licence) · get_code("e4ac3e44dca69ef9")
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 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 torch.nn.functional as F

torch.manual_seed(0)

# Set up parameters for load_balancing_loss_func
# batch_size=2, sequence_length=3, num_experts=4, num_hidden_layers=2, top_k=2

batch_size = 2
sequence_length = 3
num_experts = 4
num_hidden_layers = 2
top_k = 2

# Create gate_logits as a tuple of tensors (one per hidden layer)
# Each tensor shape: [batch_size * sequence_length, num_experts]
gate_logits = tuple(
    torch.randn(batch_size * sequence_length, num_experts)
    for _ in range(num_hidden_layers)
)

# Create attention_mask
# Shape: [batch_size, sequence_length]
attention_mask = torch.ones(batch_size, sequence_length, dtype=torch.long)

ARGS = {
    "gate_logits": gate_logits,
    "num_experts": num_experts,
    "top_k": top_k,
    "attention_mask": attention_mask,
}
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 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
load_tokenizer Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
nicolas-bzrd/llm-recipes/models/models_utils.py
pointer only (licence: NONE) · get_code("15baf103a0b4554f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

pack_pass Ran
ran on a synthesized input
arcee-ai/distillkit/distillkit/pack_logits.py
code served (permissive licence) · get_code("25fbb78bc7edcf3d")
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

torch.manual_seed(0)

# Construct test data for pack_pass
# The function expects data: dict[str, list] where each value is a list of sequences
# and max_len: int

# Create sample input_ids sequences of varying lengths
input_ids_1 = [1, 2, 3, 4, 5]
input_ids_2 = [6, 7, 8]
input_ids_3 = [9, 10, 11, 12, 13, 14, 15]
input_ids_4 = [16, 17]

# Create corresponding token_type_ids and other fields
token_type_ids_1 = [0, 0, 0, 0, 0]
token_type_ids_2 = [0, 0, 0]
token_type_ids_3 = [0, 0, 0, 0, 0, 0, 0]
token_type_ids_4 = [0, 0]

data = {
    'input_ids': [input_ids_1, input_ids_2, input_ids_3, input_ids_4],
    'token_type_ids': [token_type_ids_1, token_type_ids_2, token_type_ids_3, token_type_ids_4]
}

max_len = 10

ARGS = {
    'data': data,
    'max_len': max_len
}
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
preprocess_distillation_batch Ran
ran; our contract draft was wrong, not the code
nicolas-bzrd/llm-recipes/models/distillation_model.py
pointer only (licence: NONE) · get_code("fe57913d413a43ca")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

sparse_hinge_loss Ran
ran on a synthesized input
arcee-ai/distillkit/distillkit/lossfuncs/hingeloss.py
code served (permissive licence) · get_code("3233cdea71d52a18")
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 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

torch.manual_seed(0)

# Tiny batch setup: B=2 (batch), S=3 (sequence), K=4 (number of targets per position)
# Vocabulary size for logits dimension
V = 5

# Logits: (B, S, V) - raw model outputs for vocabulary
logits = torch.randn(2, 3, 5)

# Target IDs: (B, S, K) - indices into vocabulary for each position
# Values should be in range [0, V)
target_ids = torch.randint(0, 5, (2, 3, 4))

# Target values: (B, S, K) - either log-probabilities or probabilities
# Using small log-probability values
target_values = torch.randn(2, 3, 4) - 2  # biased toward negative for log space

# Optional mask: (B, S) - mask for valid positions in sequence
mask = torch.ones(2, 3)

# eps: small value for numerical stability
eps = 1e-08

# log_target: whether target_values are in log space
log_target = True

# margin: optional fixed margin (None means use teacher probability differences)
margin = None

ARGS = {
    "logits": logits,
    "target_ids": target_ids,
    "target_values": target_values,
    "mask": mask,
    "eps": eps,
    "log_target": log_target,
    "margin": margin,
}
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 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
sparse_logistic_ranking_loss Ran
ran on a synthesized input
arcee-ai/distillkit/distillkit/lossfuncs/logistic_ranking.py
code served (permissive licence) · get_code("a74d222992a1c151")
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 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 torch.nn.functional as F

torch.manual_seed(0)

batch_size = 2
seq_len = 3
vocab_size = 8
k_sparse = 4

# student_logits: [batch_size, seq_len, vocab_size]
student_logits = torch.randn(batch_size, seq_len, vocab_size)

# teacher_target_ids: [batch_size, seq_len, k_sparse]
# Valid indices in range [0, vocab_size)
teacher_target_ids = torch.randint(0, vocab_size, (batch_size, seq_len, k_sparse))

# teacher_target_values: [batch_size, seq_len, k_sparse]
# Log-probabilities (since log_target=True by default)
teacher_target_values = torch.randn(batch_size, seq_len, k_sparse) - 2  # negative log-probs

# log_target: boolean flag indicating values are log-probabilities
log_target = True

# sequence_mask: [batch_size, seq_len] boolean mask (True = active, False = padded)
sequence_mask = torch.ones(batch_size, seq_len, dtype=torch.bool)
sequence_mask[0, -1] = False  # mask out last position of first batch

# eps: small numerical constant
eps = 1e-9

ARGS = {
    "student_logits": student_logits,
    "teacher_target_ids": teacher_target_ids,
    "teacher_target_values": teacher_target_values,
    "log_target": log_target,
    "sequence_mask": sequence_mask,
    "eps": eps,
}
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 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
accumulate_over_chunks Not yet run
unverified (harvested, no recorded run)
arcee-ai/distillkit/distillkit/lossfuncs/common.py
code served (permissive licence) · get_code("a3052fd12526cf03")
How it ran

no invocation recorded.

dense_kl_div Not yet run
unverified (harvested, no recorded run)
arcee-ai/distillkit/distillkit/lossfuncs/kl.py
code served (permissive licence) · get_code("42f04b9fed44080e")
How it ran

no invocation recorded.

iterative_packing Not yet run
unverified (harvested, no recorded run)
arcee-ai/distillkit/distillkit/pack_logits.py
code served (permissive licence) · get_code("a8a2fd9a1ec17378")
How it ran

no invocation recorded.

load_model_from_config Not yet run
unverified (harvested, no recorded run)
nicolas-bzrd/llm-recipes/models/checkpoint_converter_fsdp_hf.py
pointer only (licence: NONE) · get_code("d96106f713c1d581")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

load_sharded_model_single_gpu Not yet run
unverified (harvested, no recorded run)
nicolas-bzrd/llm-recipes/models/checkpoint_handler.py
pointer only (licence: NONE) · get_code("93d8db92501a2a47")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, nicolas-bzrd/llm-recipes when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Also present in other papers' code

Of 19 samples lifted from this paper's linked repositories, 19 carry a function identity and 0 do not. Of the 19 distinct functions with an identity, 2 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 1 the same month, 2 later) and 17 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
load_sharded_model_single_gpu 4 other papers
2 dated earlier, 1 the same month, 1 later
2310.02469 (2023-10)
PrivacyMind: Large Language Models Can Be Contextual Privacy Protection Learners
2310.08491 (2023-10)
Prometheus: Inducing Fine-grained Evaluation Capability in Language Models
2402.18540 (2024-02)
Keeping LLMs Aligned After Fine-tuning: The Crucial Role of Prompt Templates
and 1 more not named here
get_max_seqlen_in_batch 2 other papers
2 later
2408.15793 (2024-08)
Language Adaptation on a Tight Academic Compute Budget: Tokenizer Swapping Works and Pure bfloat16 Is Enough
2501.04961 (2025-01)
Demystifying Domain-adaptive Post-training for Financial LLMs

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

Deploying large language models (LLMs) of several billion parameters can be impractical in most industrial use cases due to constraints such as cost, latency limitations, and hardware accessibility. Knowledge distillation (KD) offers a solution by compressing knowledge from resource-intensive large models to smaller ones. Various strategies exist, some relying on the text generated by the teacher model and optionally utilizing his logits to enhance learning. However, these methods based on logits often require both teacher and student models to share the same tokenizer, limiting their applicability across different LLM families. In this paper, we introduce Universal Logit Distillation (ULD) loss, grounded in optimal transport, to address this limitation. Our experimental results demonstrate the effectiveness of ULD loss in enabling distillation across models with different architectures and tokenizers, paving the way to a more widespread use of distillation techniques.

For agents

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

get_harvested_code_for_paper("2402.12030")
get_code_for_paper("2402.12030")
have("2402.12030")

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

curl https://syntology.ai/api/ran/2402.12030.json

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

[![Syntology run record](https://syntology.ai/api/ran/2402.12030.svg)](https://syntology.ai/paper/2402.12030)

Connect an agent — have() is free.