SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.12208 · CVPR · 2023

MAGVLT: Masked Generative Vision-and-Language Transformer

Sungwoong Kim, Jongmin Kim, Donghoon Lee, Jo Daejin

arXiv · PDF · Open in the Atlas

Code that ran

7 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (5 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 7 have no contract verdict; 0 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/2303.12208.json · badge: /api/ran/2303.12208.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
kakaobrain/magvlt marked official in Papers with Code
Papers with Code archive
7 ran · 3 no recorded run
FunctionStatusWhere it lives
Block Ran
ran on a synthesized input · constructed an object; no result was computed
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("89844ef2ac1027a2")
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 an instance of its own class, Block: the run constructed the object and did not call it, so no result was computed. 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 as nn
import math

torch.manual_seed(0)

# Define parameters for Block initialization
ARGS = {
    "ctx_len": 8,
    "embed_dim": 4,
    "n_heads": 2,
    "mlp_bias": True,
    "attn_bias": True,
    "resid_pdrop": 0.1,
    "attn_pdrop": 0.1,
    "gelu_use_approx": False,
}
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 Block — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Block 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
GELU Ran
ran on a synthesized input · constructed an object; no result was computed
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("0b3c58ba91896ec0")
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 an instance of its own class, GELU: the run constructed the object and did not call it, so no result was computed. 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 as nn
from torch.nn import functional as F

class GELU(nn.Module):

    def __init__(self, use_approx=False):
        super().__init__()
        self.use_approx = use_approx

    def forward(self, x):
        if self.use_approx:
            return x * torch.sigmoid(1.702 * x)
        else:
            return F.gelu(x)

torch.manual_seed(0)

# Create a small test tensor
x = torch.randn(2, 4)

ARGS = {
    "use_approx": False
}
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 GELU — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a GELU 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
LMLayer Ran
ran on a synthesized input · constructed an object; no result was computed
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("1ae0447b138f7394")
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 an instance of its own class, LMLayer: the run constructed the object and did not call it, so no result was computed. 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 as nn

torch.manual_seed(0)

# Create a small embedding module
embedding = nn.Embedding(num_embeddings=5, embedding_dim=3)

# Define ARGS dict with all parameters for LMLayer.__init__
ARGS = {
    "hidden_dim": 4,
    "intermediate_dim": 6,
    "embedding": embedding
}
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 LMLayer — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a LMLayer 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
MultiHeadSelfAttention Ran
ran on a synthesized input · constructed an object; no result was computed
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("2f73823d89add892")
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 an instance of its own class, MultiHeadSelfAttention: the run constructed the object and did not call it, so no result was computed. 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 as nn
import math
from torch.nn import functional as F

class MultiHeadSelfAttention(nn.Module):

    def __init__(self, ctx_len: int, embed_dim: int, n_heads: int, resid_pdrop: float, attn_pdrop: float, attn_bias: bool):
        super().__init__()
        assert embed_dim % n_heads == 0
        self.key = nn.Linear(embed_dim, embed_dim, bias=attn_bias)
        self.query = nn.Linear(embed_dim, embed_dim, bias=attn_bias)
        self.value = nn.Linear(embed_dim, embed_dim, bias=attn_bias)
        self.attn_drop = nn.Dropout(attn_pdrop)
        self.resid_drop = nn.Dropout(resid_pdrop)
        self.proj = nn.Linear(embed_dim, embed_dim, attn_bias)
        self.n_heads = n_heads
        self.ctx_len = ctx_len

    def forward(self, x, use_cache=False, layer_past=None, attn_mask=None):
        B, T, C = x.shape
        x = x.transpose(0, 1).contiguous()
        k = self.key(x).view(T, B * self.n_heads, C // self.n_heads).transpose(0, 1)
        q = self.query(x).view(T, B * self.n_heads, C // self.n_heads).transpose(0, 1)
        v = self.value(x).view(T, B * self.n_heads, C // self.n_heads).transpose(0, 1)
        if use_cache:
            present = torch.stack([k, v])
        if layer_past is not None:
            past_key, past_value = layer_past
            k = torch.cat([past_key, k], dim=-2)
            v = torch.cat([past_value, v], dim=-2)
        att = torch.bmm(q, k.transpose(-2, -1) * (1.0 / math.sqrt(k.size(-1))))
        if attn_mask is not None:
            no_attn_ids = torch.logical_not(attn_mask.sum(dim=-1))
            attn_mask[torch.diag_embed(no_attn_ids)] = True
            mask = torch.repeat_interleave(attn_mask, self.n_heads, dim=0)
            att = att.masked_fill(mask == 0, float('-inf'))
        att = F.softmax(att, dim=-1)
        att = self.attn_drop(att)
        y = torch.bmm(att, v)
        y = y.transpose(0, 1).contiguous().view(T, B, C)
        y = self.resid_drop(self.proj(y))
        if use_cache:
            return (y.transpose(0, 1).contiguous(), present)
        else:
            return y.transpose(0, 1).contiguous()

    def sample(self, x, layer_past=None):
        x, present = self(x, use_cache=True, layer_past=layer_past)
        return (x, present)

torch.manual_seed(0)

ctx_len = 8
embed_dim = 4
n_heads = 2
resid_pdrop = 0.1
attn_pdrop = 0.1
attn_bias = True

batch_size = 2
seq_len = 3
x = torch.randn(batch_size, seq_len, embed_dim)

ARGS = {
    "ctx_len": ctx_len,
    "embed_dim": embed_dim,
    "n_heads": n_heads,
    "resid_pdrop": resid_pdrop,
    "attn_pdrop": attn_pdrop,
    "attn_bias": attn_bias,
}
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 MultiHeadSelfAttention — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MultiHeadSelfAttention 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
PositionalEmbed Ran
ran on a synthesized input
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("08e6f64d649cf0e9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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
import math

class Sinusoidal1DPositionalEmbed(nn.Module):

    def __init__(self, length=64, embed_dim=1024, learnable=True):
        super().__init__()
        if embed_dim % 2 != 0:
            raise ValueError('Cannot use sin/cos positional encoding with odd dimension (got dim={:d})'.format(embed_dim))
        self.length = length
        self.embed_dim = embed_dim
        self.dim = self.embed_dim
        div = torch.exp(torch.arange(0.0, self.dim, 2) * -(math.log(10000.0) / self.dim))
        if learnable:
            self.div = nn.Parameter(div)
        else:
            self.register_buffer('div', div)

    def forward(self, x):
        B = x.shape[0]
        L = x.shape[1]
        pos_embed = torch.zeros(B, L, self.embed_dim).to(x.device)
        pos_embed[:, :, 0:self.dim:2] += torch.sin(x.unsqueeze(-1) * self.div)
        pos_embed[:, :, 1:self.dim:2] += torch.cos(x.unsqueeze(-1) * self.div)
        return pos_embed

class Sinusoidal2DPositionalEmbed(nn.Module):

    def __init__(self, length=256, embed_dim=1024, learnable=True):
        super().__init__()
        if embed_dim % 4 != 0:
            raise ValueError('Cannot use sin/cos positional encoding with odd dimension (got dim={:d})'.format(embed_dim))
        self.length = length
        self.HW = int(self.length ** 0.5)
        self.embed_dim = embed_dim
        self.dim = int(self.embed_dim / 2)
        div = torch.exp(torch.arange(0.0, self.dim, 2) * -(math.log(10000.0) / self.dim))
        div = div.repeat([2])
        if learnable:
            self.div = nn.Parameter(div)
        else:
            self.register_buffer('div', div)
        self.hw_idx = torch.tensor([[[h, w] for w in range(self.HW)] for h in range(self.HW)]).to(torch.float)
        self.hw_idx = self.hw_idx.reshape([self.length, 2])

    def forward(self, x):
        B = x.shape[0]
        L = x.shape[1]
        pos_w = self.hw_idx[x, 1].unsqueeze(-1).to(x.device)
        pos_h = self.hw_idx[x, 0].unsqueeze(-1).to(x.device)
        pos_embed = torch.zeros(B, L, self.embed_dim).to(x.device)
        pos_embed[:, :, 0:self.dim:2] += torch.sin(pos_w * self.div[:self.dim // 2]).reshape([B, L, -1])
        pos_embed[:, :, 1:self.dim:2] += torch.cos(pos_w * self.div[:self.dim // 2]).reshape([B, L, -1])
        pos_embed[:, :, self.dim::2] += torch.sin(pos_h * self.div[self.dim // 2:]).reshape([B, L, -1])
        pos_embed[:, :, self.dim + 1::2] += torch.cos(pos_h * self.div[self.dim // 2:]).reshape([B, L, -1])
        return pos_embed

class PositionalEmbed(nn.Module):

    def __init__(self, mode='la1d', length=64, embed_dim=1024):
        super().__init__()
        self.mode = mode
        self.length = length
        if self.mode == 'la1d':
            self.pos_embed = nn.Parameter(torch.zeros(length, embed_dim))
            self.pos_embed.data.normal_(mean=0.0, std=0.02)
        elif self.mode == 's2d':
            self.pos_embed = Sinusoidal2DPositionalEmbed(length=length, embed_dim=embed_dim, learnable=False)
        elif self.mode == 'ls2d':
            self.pos_embed = Sinusoidal2DPositionalEmbed(length=length, embed_dim=embed_dim, learnable=True)
        elif self.mode == 's1d':
            self.pos_embed = Sinusoidal1DPositionalEmbed(length=length, embed_dim=embed_dim, learnable=False)
        elif self.mode == 'ls1d':
            self.pos_embed = Sinusoidal1DPositionalEmbed(length=length, embed_dim=embed_dim, learnable=True)
        else:
            raise NotImplementedError

    def forward(self, x):
        if len(x.shape) == 1:
            x = x.unsqueeze(0)
        if self.mode == 'la1d':
            pos_embed = self.pos_embed[x, :]
        else:
            pos_embed = self.pos_embed(x)
        return pos_embed


INIT_ARGS = {
    'mode': 'la1d',
    'length': 16,
    'embed_dim': 8
}

ARGS = {
    'x': torch.tensor([[0, 1, 2, 3], [4, 5, 6, 7]], dtype=torch.long)
}

model = PositionalEmbed(**INIT_ARGS)
output = model(**ARGS)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Sinusoidal1DPositionalEmbed Ran
ran on a synthesized input
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("363e79d3badd3e32")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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
import math

class Sinusoidal1DPositionalEmbed(nn.Module):

    def __init__(self, length=64, embed_dim=1024, learnable=True):
        super().__init__()
        if embed_dim % 2 != 0:
            raise ValueError('Cannot use sin/cos positional encoding with odd dimension (got dim={:d})'.format(embed_dim))
        self.length = length
        self.embed_dim = embed_dim
        self.dim = self.embed_dim
        div = torch.exp(torch.arange(0.0, self.dim, 2) * -(math.log(10000.0) / self.dim))
        if learnable:
            self.div = nn.Parameter(div)
        else:
            self.register_buffer('div', div)

    def forward(self, x):
        B = x.shape[0]
        L = x.shape[1]
        pos_embed = torch.zeros(B, L, self.embed_dim).to(x.device)
        pos_embed[:, :, 0:self.dim:2] += torch.sin(x.unsqueeze(-1) * self.div)
        pos_embed[:, :, 1:self.dim:2] += torch.cos(x.unsqueeze(-1) * self.div)
        return pos_embed


INIT_ARGS = {
    'length': 16,
    'embed_dim': 8,
    'learnable': True
}

ARGS = {
    'x': torch.arange(0, 3, dtype=torch.float32).unsqueeze(0).expand(2, -1)
}

if __name__ == '__main__':
    model = Sinusoidal1DPositionalEmbed(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Sinusoidal2DPositionalEmbed Ran
ran on a synthesized input · constructed an object; no result was computed
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("1bad1f729cc2768c")
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 an instance of its own class, Sinusoidal2DPositionalEmbed: the run constructed the object and did not call it, so no result was computed. 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 as nn
import math

torch.manual_seed(0)

# Construct arguments for Sinusoidal2DPositionalEmbed
# The forward method expects:
# - x: a tensor of shape (B, L) containing indices into hw_idx
# - The module needs length, embed_dim, and learnable parameters

# Create a small valid instance
# length must be a perfect square for self.HW = int(self.length ** 0.5)
# embed_dim must be divisible by 4

length = 16  # 4x4 grid
embed_dim = 8  # divisible by 4
learnable = True

ARGS = {
    'length': length,
    'embed_dim': embed_dim,
    'learnable': learnable,
}
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 Sinusoidal2DPositionalEmbed — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Sinusoidal2DPositionalEmbed 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
ModelTemplate Not yet run
unverified (harvested, no recorded run)
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("94eaf59881a7f6f3")
How it ran

no invocation recorded.

SPECIAL_TOKENS Not yet run
unverified (harvested, no recorded run)
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("53ebad6e00fab368")
How it ran

no invocation recorded.

Transformer1d Not yet run
unverified (harvested, no recorded run)
kakaobrain/magvlt/magvlt/models/stage2/transformers.py
code served (permissive licence) · get_code("6edc6c8c22bba74b")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 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.

FunctionAlso present inEarliest-datedOthers named
GELU
brick of this paper's assembly: helper
1 other paper
1 dated earlier · 1 only as an assembly brick
2209.06192 (2022-09)
StoryDALL-E: Adapting Pretrained Text-to-Image Transformers for Story Continuation

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

While generative modeling on multimodal image-text data has been actively developed with large-scale paired datasets, there have been limited attempts to generate both image and text data by a single model rather than a generation of one fixed modality conditioned on the other modality. In this paper, we explore a unified generative visionand-language (VL) model that can produce both images and text sequences. Especially, we propose a generative VL transformer based on the non-autoregressive mask prediction, named MAGVLT, and compare it with an autoregressive generative VL transformer (ARGVLT). In comparison to ARGVLT, the proposed MAGVLT enables bidirectional context encoding, fast decoding by parallel token predictions in an iterative refinement, and extended editing capabilities such as image and text infilling. For rigorous training of our MAGVLT with image-text pairs from scratch, we combine the image-to-text, text-to-image, and joint imageand-text mask prediction tasks. Moreover, we devise two additional tasks based on the step-unrolled mask prediction and the selective prediction on the mixture of two image-text pairs. Experimental results on various downstream generation tasks of VL benchmarks show that our MAGVLT outperforms ARGVLT by a large margin even with significant inference speedup. Particularly, MAGVLT achieves competitive results on both zero-shot image-to-text and text-toimage generation tasks from MS-COCO by one moderatesized model (fewer than 500M parameters) even without the use of monomodal data and networks.

For agents

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

get_harvested_code_for_paper("2303.12208")
get_code_for_paper("2303.12208")
have("2303.12208")

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

curl https://syntology.ai/api/ran/2303.12208.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.12208.svg)](https://syntology.ai/paper/2303.12208)

Connect an agent — have() is free.