SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.12705 · 2024

WorldCuisines: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines

arXiv · PDF · Open in the Atlas

Code that ran

9 of 17 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 1 has no contract verdict; 8 failures of our instrument; 8 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/2410.12705.json · badge: /api/ran/2410.12705.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
worldcuisines/worldcuisines canonical (classifier label)
classifier basis: census
9 ran · 8 no recorded run
FunctionStatusWhere it lives
eval_instance Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/src/aria.py
code served (permissive licence) · get_code("c4227ba3d818e9d5")
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. 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
import torch.nn as nn
from PIL import Image
import io

torch.manual_seed(0)

# Stand-in model that mimics the expected interface
class MockVisionLanguageModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.embedding = nn.Embedding(1000, 8)
        self.linear = nn.Linear(8, 8)
        self.dtype = torch.float32
        self.device = torch.device('cpu')
    
    def generate(self, input_ids, pixel_values, attention_mask=None, max_new_tokens=50, stop_strings=None, tokenizer=None, do_sample=True, top_k=1):
        # Return dummy token ids: original input_ids plus some generated tokens
        batch_size = input_ids.shape[0]
        generated = torch.randint(0, 1000, (batch_size, max_new_tokens), dtype=torch.long)
        return torch.cat([input_ids, generated], dim=1)

# Stand-in processor
class MockProcessor:
    def __init__(self):
        self.tokenizer = MockTokenizer()
    
    def apply_chat_template(self, conversation, add_generation_prompt=True):
        # Return a dummy prompt string
        return "This is a test prompt with image and text"
    
    def __call__(self, images=None, text=None, return_tensors='pt'):
        # Return dummy tensors
        batch_size = 1
        seq_length = 10
        num_patches = 64
        
        inputs = {
            'input_ids': torch.randint(0, 1000, (batch_size, seq_length), dtype=torch.long),
            'attention_mask': torch.ones((batch_size, seq_length), dtype=torch.long),
            'pixel_values': torch.randn((batch_size, 3, 224, 224), dtype=torch.float32),
        }
        return inputs
    
    def decode(self, token_ids, skip_special_tokens=True):
        # Return dummy decoded text
        return "This is a test response"

# Stand-in tokenizer
class MockTokenizer:
    def __init__(self):
        pass

# Create tiny PIL image
def create_test_image():
    img = Image.new('RGB', (224, 224), color='red')
    return img

model = MockVisionLanguageModel()
processor = MockProcessor()
image_file = create_test_image()
query = "What is in this image?"

ARGS = {
    'model': model,
    'processor': processor,
    'image_file': image_file,
    'query': query,
    'tokenizer': None,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, 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
eval_instance Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/src/llama.py
code served (permissive licence) · get_code("884a4e29f634f7da")
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. 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
import torch.nn as nn
from PIL import Image
import io

torch.manual_seed(0)

class MockProcessor:
    def apply_chat_template(self, conversation, add_generation_prompt=False):
        return "mock prompt"
    
    def __call__(self, images, text, return_tensors=None):
        batch_size = 1 if not isinstance(images, list) else len(images)
        mock_output = {
            'input_ids': torch.tensor([[1, 2, 3, 4, 5]], dtype=torch.long),
            'pixel_values': torch.randn(batch_size, 3, 224, 224),
        }
        return MockTensorDict(mock_output)
    
    def decode(self, token_ids, skip_special_tokens=False):
        return "generated response text"

class MockTensorDict(dict):
    def to(self, device):
        return self

class MockModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def generate(self, input_ids, pixel_values, max_new_tokens=50, top_k=1, **kwargs):
        batch_size = input_ids.size(0)
        seq_len = input_ids.size(1) + max_new_tokens
        output = torch.randint(0, 100, (batch_size, seq_len), dtype=torch.long)
        return output

def create_mock_image():
    img = Image.new('RGB', (224, 224), color='red')
    return img

model = MockModel()
processor = MockProcessor()
image_file = create_mock_image()
query = "What is in this image?"
tokenizer = None

ARGS = {
    'model': model,
    'processor': processor,
    'image_file': image_file,
    'query': query,
    'tokenizer': tokenizer,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, 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
eval_instance Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/src/llava.py
code served (permissive licence) · get_code("ef2279719389627b")
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. 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
import torch.nn as nn
from PIL import Image
import io

torch.manual_seed(0)

class MockProcessor:
    def apply_chat_template(self, conversation, add_generation_prompt=False):
        return "User: test query ASSISTANT: "
    
    def __call__(self, images=None, text=None, return_tensors=None):
        # Return a dict with 'input_ids' and 'pixel_values' keys
        result = {
            'input_ids': torch.randint(0, 100, (1, 10)),
            'pixel_values': torch.randn(1, 3, 224, 224),
            'attention_mask': torch.ones(1, 10)
        }
        # Mock the .to() method by returning self with modified tensors
        return MockTensorDict(result)
    
    def decode(self, tensor, skip_special_tokens=False):
        return "User: test query ASSISTANT: This is a test response from the model."

class MockTensorDict:
    def __init__(self, data):
        self.data = data
    
    def to(self, device):
        return self
    
    def keys(self):
        return self.data.keys()
    
    def __getitem__(self, key):
        return self.data[key]
    
    def __iter__(self):
        return iter(self.data)

class MockModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 4)
    
    def generate(self, **kwargs):
        # Return a tensor of shape (1, seq_len) with token ids
        return torch.randint(0, 100, (1, 20))

def create_dummy_image():
    # Create a tiny PIL Image for testing
    img_array = torch.randint(0, 256, (3, 32, 32), dtype=torch.uint8)
    # Convert to PIL Image
    img_array_np = img_array.numpy()
    img = Image.fromarray(img_array_np.transpose(1, 2, 0))
    return img

model = MockModel()
processor = MockProcessor()
image_file = create_dummy_image()
query = "What is in this image?"
tokenizer = None

ARGS = {
    'model': model,
    'processor': processor,
    'image_file': image_file,
    'query': query,
    'tokenizer': tokenizer
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, 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
final_cat_mapping Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/score/score.py
code served (permissive licence) · get_code("8fb75acb8c7ec039")
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.

from collections import OrderedDict

torch_imported = True

# Construct a valid scores dictionary
# The function expects a dict where values can be dicts with specific keys
scores = {
    '1': {
        'su_loma': 0.85,
        'si_formal_spoken': 0.92,
        'other_metric': 0.78
    },
    '2': {
        'su_loma': 0.81,
        'si_formal_spoken': 0.88,
        'another_metric': 0.75
    },
    '3': {
        'metric_a': 0.90,
        'metric_b': 0.87
    },
    '4': 0.79
}

ARGS = {
    'scores': scores
}
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 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
get_unique_filename Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/src/base_model.py
code served (permissive licence) · get_code("92244c7286cce0f9")
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 os
import tempfile

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

# Test case 1: File that doesn't exist yet
test_path = os.path.join(temp_dir, "test_file.txt")

ARGS = {
    "path": test_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 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
get_url_jpg_map Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/src/base_model.py
code served (permissive licence) · get_code("c5c29786001f4200")
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 pandas as pd
from PIL import Image
from io import BytesIO
import numpy as np

# Create a small PIL Image
def create_test_image():
    img = Image.new('RGB', (10, 10), color='red')
    buffer = BytesIO()
    img.save(buffer, format='PNG')
    buffer.seek(0)
    return buffer.getvalue()

# Create test data as a DataFrame
image_data = []
for row_idx in range(2):
    row_dict = {}
    for i in range(1, 9):
        url = f"http://example.com/image{i}_{row_idx}.jpg"
        image_bytes = create_test_image()
        row_dict[f'image{i}_url'] = url
        row_dict[f'image{i}'] = {'bytes': image_bytes}
    image_data.append(row_dict)

data = pd.DataFrame(image_data)

ARGS = {
    'data': data
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, pandas — 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
load_jsonl Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/score/score.py
code served (permissive licence) · get_code("bf44cfc1771f959a")
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 json
import tempfile
import os

# Create a temporary JSONL file with valid test data
temp_dir = tempfile.mkdtemp()
temp_file = os.path.join(temp_dir, "test_data.jsonl")

# Write test data to the file
with open(temp_file, 'w', encoding='utf-8') as f:
    # Write valid JSON lines
    f.write(json.dumps({"key": "value1", "number": 42}) + "\n")
    f.write(json.dumps({"key": "value2", "number": 99}) + "\n")
    f.write(json.dumps({"nested": {"inner": "data"}}) + "\n")
    # Write a line that looks like Python literal (for ast.literal_eval fallback)
    f.write("{'key': 'value3', 'number': 123}\n")

ARGS = {
    "file_path": 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
load_model_processor Ran
ran on a synthesized input
worldcuisines/worldcuisines/evaluation/src/molmo.py
code served (permissive licence) · get_code("1af4a289b725ad56")
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 2, dtype family float; no output shape is stated. 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 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 MagicMock, patch
import sys

torch.manual_seed(0)

# Create mock for AutoProcessor
mock_processor = MagicMock()
mock_processor.from_pretrained = MagicMock(return_value=MagicMock())

# Create mock for AutoModelForCausalLM
mock_model = MagicMock()
mock_model.from_pretrained = MagicMock(return_value=MagicMock())

# Mock the transformers module components
sys.modules['transformers'] = MagicMock()
sys.modules['transformers'].AutoProcessor = MagicMock(from_pretrained=lambda *args, **kwargs: MagicMock())
sys.modules['transformers'].AutoModelForCausalLM = MagicMock(from_pretrained=lambda *args, **kwargs: MagicMock())
sys.modules['transformers'].GenerationConfig = MagicMock()

ARGS = {
    'model_path': 'gpt2',
    'fp32': False,
    'multi_gpu': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, transformers — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, 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
load_uploaded_files Ran
ran; our contract draft was wrong, not the code
worldcuisines/worldcuisines/evaluation/gemini.py
code served (permissive licence) · get_code("3acf8494948af533")
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 os
import json
import tempfile

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

# Create a sample uploaded_files JSON file
uploaded_files_data = {
    "file1.txt": "/path/to/file1.txt",
    "file2.pdf": "/path/to/file2.pdf",
    "file3.csv": "/path/to/file3.csv"
}

uploaded_files_file = os.path.join(temp_dir, "uploaded_files.json")
with open(uploaded_files_file, 'w') as f:
    json.dump(uploaded_files_data, f)

ARGS = {
    "uploaded_files_file": uploaded_files_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 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
eval_instance Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/src/molmo.py
code served (permissive licence) · get_code("4423a1422aa933d1")
How it ran

no invocation recorded.

get_image Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/gemini.py
code served (permissive licence) · get_code("59daa7bc4cd39ffa")
How it ran

no invocation recorded.

get_vqa_from_hf Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/generate_openai_batch_api.py
code served (permissive licence) · get_code("ad4fdc995ecb91f4")
How it ran

no invocation recorded.

get_vqa_from_hf Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/src/base_model.py
code served (permissive licence) · get_code("e57f7366380e2537")
How it ran

no invocation recorded.

load_model_processor Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/src/aria.py
code served (permissive licence) · get_code("cb78c8e64d7b0e3a")
How it ran

no invocation recorded.

load_model_processor Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/src/llama.py
code served (permissive licence) · get_code("d3cf62ea670a86f0")
How it ran

no invocation recorded.

load_model_processor Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/src/llava.py
code served (permissive licence) · get_code("b981a004699d96c8")
How it ran

no invocation recorded.

load_result Not yet run
unverified (harvested, no recorded run)
worldcuisines/worldcuisines/evaluation/score/score.py
code served (permissive licence) · get_code("d2fb3f56c8a5c3b4")
How it ran

no invocation recorded.

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 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 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.

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

Vision Language Models (VLMs) often struggle with culture-specific knowledge, particularly in languages other than English and in underrepresented cultural contexts. To evaluate their understanding of such knowledge, we introduce WorldCuisines, a massive-scale benchmark for multilingual and multicultural, visually grounded language understanding. This benchmark includes a visual question answering (VQA) dataset with text-image pairs across 30 languages and dialects, spanning 9 language families and featuring over 1 million data points, making it the largest multicultural VQA benchmark to date. It includes tasks for identifying dish names and their origins. We provide evaluation datasets in two sizes (12k and 60k instances) alongside a training dataset (1 million instances). Our findings show that while VLMs perform better with correct location context, they struggle with adversarial contexts and predicting specific regional cuisines and languages. To support future research, we release a knowledge base with annotated food entries and images along with the VQA data.

For agents

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

get_harvested_code_for_paper("2410.12705")
get_code_for_paper("2410.12705")
have("2410.12705")

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

curl https://syntology.ai/api/ran/2410.12705.json

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

[![Syntology run record](https://syntology.ai/api/ran/2410.12705.svg)](https://syntology.ai/paper/2410.12705)

Connect an agent — have() is free.