SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.21054 · 2026

VAUQ: Vision-Aware Uncertainty Quantification for LVLM Self-Evaluation

Changdae Oh, Hyeong Choi, Sharon Li, Seongheon Park, Sean Du

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 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; 3 failures of our instrument; 0 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.21054.json · badge: /api/ran/2602.21054.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
deeplearning-wisc/vauq community
Papers with Code archive
4 ran · 0 no recorded run
FunctionStatusWhere it lives
OutputScoreInfo Ran
ran on a synthesized input · constructed an object; no result was computed
deeplearning-wisc/vauq/vauq/scoring.py
code served (permissive licence) · get_code("0ca9b0d0bc506b96")
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, OutputScoreInfo: 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 numpy as np
import torch.nn.functional as F
from scipy.stats import entropy

torch.manual_seed(0)

# Create small tensors for the function
batch_size = 2
vocab_size = 4
seq_length = 3

# response_logits: shape (seq_length, vocab_size) - logits for each token
response_logits = torch.randn(seq_length, vocab_size, dtype=torch.float32)

# response_ids: tensor of token ids, shape (seq_length,)
response_ids = torch.randint(0, vocab_size, (seq_length,), dtype=torch.long)

# device: CPU device
device = torch.device('cpu')

ARGS = {
    'response_logits': response_logits,
    'response_ids': response_ids,
    'device': device
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type OutputScoreInfo — 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 OutputScoreInfo 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
compute_entropy Ran
ran; our contract draft was wrong, not the code
deeplearning-wisc/vauq/vauq/scoring.py
code served (permissive licence) · get_code("d1f6db28a2a7ff6a")
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 torch.nn as nn
import numpy as np

torch.manual_seed(0)

class MockLVLM(nn.Module):
    """Minimal mock LVLM that implements get_logits and has a device attribute."""
    
    def __init__(self, vocab_size=100, device='cpu'):
        super().__init__()
        self.device = device
        self.vocab_size = vocab_size
        # Small linear layer to generate logits
        self.linear = nn.Linear(16, vocab_size)
    
    def get_logits(self, image, question, generated_ids):
        """
        Mock implementation that returns logits.
        Returns shape (seq_len, vocab_size) where seq_len matches generated_ids length.
        """
        seq_len = generated_ids.shape[0]
        # Create dummy logits of shape (seq_len, vocab_size)
        dummy_input = torch.randn(seq_len, 16, device=self.device)
        logits = self.linear(dummy_input)
        return logits

# Create mock inputs
vocab_size = 100
device = 'cpu'
lvlm = MockLVLM(vocab_size=vocab_size, device=device)

# Mock image (can be any object, not actually used in compute_entropy)
image = torch.randn(3, 32, 32)

# Mock question (can be any object, not actually used)
question = "What is this?"

# Mock generated_ids - sequence of token ids
# Shape: (seq_len,) - let's use seq_len=5
generated_ids = torch.tensor([1, 5, 10, 42, 7], dtype=torch.long, device=device)

ARGS = {
    'lvlm': lvlm,
    'image': image,
    'question': question,
    'generated_ids': generated_ids
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, 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
compute_entropy_core_masked Ran
ran; our contract draft was wrong, not the code
deeplearning-wisc/vauq/vauq/scoring.py
code served (permissive licence) · get_code("56dfc7e0b1b512fd")
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 torch.nn as nn
import numpy as np

torch.manual_seed(0)

class MockLVLM(nn.Module):
    """Mock LVLM model with minimal interface for the function."""
    
    def __init__(self):
        super().__init__()
        self.device = torch.device('cpu')
        self.linear = nn.Linear(8, 256)  # vocab size 256
    
    def get_logits_masked(self, image, question, generated_ids, topk_ratio=0.5, layer_range=(0, 12)):
        """
        Returns logits of shape (seq_len, vocab_size) where seq_len matches generated_ids length.
        """
        seq_len = len(generated_ids)
        # Return logits: (seq_len, vocab_size)
        logits = torch.randn(seq_len, 256, device=self.device)
        return logits

# Create mock LVLM
mock_lvlm = MockLVLM()

# Create mock inputs
mock_image = torch.randn(3, 32, 32)  # Simple image tensor
mock_question = "What is this?"  # String question
mock_generated_ids = torch.tensor([1, 2, 3, 4, 5], dtype=torch.long)  # Token IDs, length 5

# Parameters for the function
topk_ratio = 0.5
layer_range = [0, 12]

ARGS = {
    'lvlm': mock_lvlm,
    'image': mock_image,
    'question': mock_question,
    'generated_ids': mock_generated_ids,
    'topk_ratio': topk_ratio,
    'layer_range': layer_range
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, 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
compute_vauq_scores Ran
ran; our contract draft was wrong, not the code
deeplearning-wisc/vauq/vauq/scoring.py
code served (permissive licence) · get_code("db7a1d580b6d1586")
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 torch.nn as nn
import torch.nn.functional as F
import numpy as np

torch.manual_seed(0)

class MockLVLM(nn.Module):
    """Mock vision-language model that implements required methods."""
    
    def __init__(self, vocab_size=100, hidden_dim=16):
        super().__init__()
        self.vocab_size = vocab_size
        self.hidden_dim = hidden_dim
        self.device = torch.device('cpu')
        self.embed = nn.Embedding(vocab_size, hidden_dim)
        self.linear = nn.Linear(hidden_dim, vocab_size)
        
    def get_logits(self, image, question, generated_ids):
        """Returns logits for generated_ids tokens."""
        # Simulate logits: [seq_len, vocab_size]
        seq_len = generated_ids.shape[0] if generated_ids.dim() > 0 else 1
        logits = torch.randn(seq_len, self.vocab_size, dtype=torch.float32)
        return logits
    
    def get_logits_masked(self, image, question, generated_ids, topk_ratio=0.6, layer_range=(10, 25)):
        """Returns logits with core masking applied."""
        seq_len = generated_ids.shape[0] if generated_ids.dim() > 0 else 1
        logits = torch.randn(seq_len, self.vocab_size, dtype=torch.float32)
        return logits


# Create mock LVLM instance
lvlm = MockLVLM(vocab_size=100, hidden_dim=16)

# Create mock image (any tensor works, just needs to be passable)
image = torch.randn(3, 32, 32)

# Create mock question (any tensor works)
question = torch.tensor([1, 2, 3, 4])

# Create mock generated_ids (sequence of token IDs)
generated_ids = torch.tensor([5, 10, 15, 20, 25])

# Define all arguments for compute_vauq_scores
ARGS = {
    'lvlm': lvlm,
    'image': image,
    'question': question,
    'generated_ids': generated_ids,
    'topk_ratio': 0.6,
    'alpha': 0.5,
    'layer_range': (10, 25)
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, scipy, 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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 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 4 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

Large Vision-Language Models (LVLMs) frequently hallucinate, limiting their safe deployment in real-world applications. Existing LLM self-evaluation methods rely on a model's ability to estimate the correctness of its own outputs, which can improve deployment reliability; however, they depend heavily on language priors and are therefore ill-suited for evaluating vision-conditioned predictions. We propose VAUQ, a vision-aware uncertainty quantification framework for LVLM self-evaluation that explicitly measures how strongly a model's output depends on visual evidence. VAUQ introduces the Image-Information Score (IS), which captures the reduction in predictive uncertainty attributable to visual input, and an unsupervised core-region masking strategy that amplifies the influence of salient regions. Combining predictive entropy with this core-masked IS yields a training-free scoring function that reliably reflects answer correctness. Comprehensive experiments show that VAUQ consistently outperforms existing self-evaluation methods across multiple datasets 1 .

For agents

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

get_harvested_code_for_paper("2602.21054")
get_code_for_paper("2602.21054")
have("2602.21054")

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

curl https://syntology.ai/api/ran/2602.21054.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.21054.svg)](https://syntology.ai/paper/2602.21054)

Connect an agent — have() is free.