SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2501.08282 · CVPR · 2025

LLaVA-ST: A Multimodal Large Language Model for Fine-Grained Spatial-Temporal Understanding

Si Liu, Hongyu Li, Jialin Gao, Xiaoming Wei, Shaofei Huang, Tianrui Hui, Jinyu Chen, Ziyu Wei

arXiv · PDF · Open in the Atlas

Code that ran

3 of 7 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 2 have no contract verdict; 1 failure of our instrument; 4 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/2501.08282.json · badge: /api/ran/2501.08282.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
appletea233/LLaVA-ST community
Papers with Code archive
3 ran · 4 no recorded run
FunctionStatusWhere it lives
HFVisionTower Ran
ran on a synthesized input · constructed an object; no result was computed
appletea233/LLaVA-ST/llava/model/llava_arch.py
code served (permissive licence) · get_code("4b28bde89dfaf861")
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, HFVisionTower: the run constructed the object and did not call it, so no result was computed. 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 unittest.mock import Mock, MagicMock

torch.manual_seed(0)

# Create a minimal mock args object
class Args:
    def __init__(self):
        self.mm_vision_select_layer = -2
        self.mm_vision_select_feature = 'patch'

# Create a minimal mock vision tower config
class MockVisionConfig:
    def __init__(self):
        self.hidden_size = 768
        self.image_size = 224
        self.patch_size = 16
        self.vision_config = self

# Create a minimal mock vision model
class MockVisionModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.config = MockVisionConfig()
        self.device = torch.device('cpu')
        self.dtype = torch.float32
        
    def forward(self, x, output_hidden_states=False):
        # Return mock output with hidden_states
        batch_size = x.shape[0]
        num_patches = (self.config.image_size // self.config.patch_size) ** 2
        hidden_size = self.config.hidden_size
        
        # Create mock hidden states (list of tensors from different layers)
        hidden_states = [
            torch.randn(batch_size, num_patches + 1, hidden_size),
            torch.randn(batch_size, num_patches + 1, hidden_size),
            torch.randn(batch_size, num_patches + 1, hidden_size),
            torch.randn(batch_size, num_patches + 1, hidden_size),
        ]
        
        output = Mock()
        output.hidden_states = hidden_states
        return output
    
    def requires_grad_(self, requires_grad):
        for param in self.parameters():
            param.requires_grad = requires_grad
        return self
    
    def to(self, device):
        return self

# Create a minimal mock image processor
class MockImageProcessor:
    def __init__(self):
        pass

# Patch AutoModel and AutoImageProcessor
import transformers
original_auto_model = transformers.AutoModel.from_pretrained
original_auto_processor = transformers.AutoImageProcessor.from_pretrained
original_auto_config = transformers.AutoConfig.from_pretrained

def mock_from_pretrained_model(model_name, **kwargs):
    return MockVisionModel()

def mock_from_pretrained_processor(model_name, **kwargs):
    return MockImageProcessor()

def mock_from_pretrained_config(model_name, **kwargs):
    return MockVisionConfig()

transformers.AutoModel.from_pretrained = mock_from_pretrained_model
transformers.AutoImageProcessor.from_pretrained = mock_from_pretrained_processor
transformers.AutoConfig.from_pretrained = mock_from_pretrained_config

ARGS = {
    'vision_tower': 'hf:openai/clip-vit-large-patch14',
    'args': Args(),
    'delay_load': 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 HFVisionTower — 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 HFVisionTower 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
OpenCLIPVisionTower Ran
ran on a synthesized input · constructed an object; no result was computed
appletea233/LLaVA-ST/llava/model/llava_arch.py
code served (permissive licence) · get_code("6ccef5c9c25d4935")
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, OpenCLIPVisionTower: the run constructed the object and did not call it, so no result was computed. 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

torch.manual_seed(0)

class MockArgs:
    def __init__(self):
        self.vision_tower_pretrained = 'openai'
        self.mm_vision_select_layer = 0
        self.mm_vision_select_feature = 'patch'
        self.unfreeze_mm_vision_tower = False
        self.mm_tunable_parts = []

class MockImageProcessor:
    def __init__(self):
        self.transforms = []

class MockConv1:
    def __init__(self):
        self.kernel_size = (14,)
        self.weight = torch.randn(768, 3, 14, 14)

class MockTransformer:
    def __init__(self):
        self.resblocks = []

class MockVisionTower(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = MockConv1()
        self.class_embedding = torch.randn(1, 768)
        self.positional_embedding = torch.randn(50, 768)
        self.patch_dropout = nn.Identity()
        self.ln_pre = nn.LayerNorm(768)
        self.transformer = MockTransformer()
    
    def forward(self, x):
        return x

ARGS = {
    'vision_tower': 'open_clip_hub:ViT-H-14-378-quickgelu',
    'args': MockArgs(),
    'delay_load': True
}
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 OpenCLIPVisionTower — 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 OpenCLIPVisionTower 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
build_vision_resampler Ran
ran; our contract draft was wrong, not the code
appletea233/LLaVA-ST/llava/model/llava_arch.py
code served (permissive licence) · get_code("8b328600dceb2a03")
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 IdentityMap. 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)

class MockModelArgs:
    def __init__(self):
        self.mm_resampler_type = None
        self.mm_vision_tower = None
        self.vision_tower = None

ARGS = {
    'model_args': MockModelArgs(),
    'delay_load': 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 IdentityMap — a IdentityMap output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a IdentityMap 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
ImageBindWrapper Not yet run
unverified (harvested, no recorded run)
appletea233/LLaVA-ST/llava/model/llava_arch.py
code served (permissive licence) · get_code("ed8d985533703c13")
How it ran

no invocation recorded.

LlavaMetaModel Not yet run
unverified (harvested, no recorded run)
appletea233/LLaVA-ST/llava/model/llava_arch.py
code served (permissive licence) · get_code("a59d94054d821180")
How it ran

no invocation recorded.

VisionConfig Not yet run
unverified (harvested, no recorded run)
appletea233/LLaVA-ST/llava/model/llava_arch.py
code served (permissive licence) · get_code("7d5165195a4a723a")
How it ran

no invocation recorded.

build_vision_tower Not yet run
unverified (harvested, no recorded run)
appletea233/LLaVA-ST/llava/model/llava_arch.py
code served (permissive licence) · get_code("e081f77558b6872a")
How it ran

no invocation recorded.

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 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 7 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

The event duration is <16.2s-22s>, at 16.2s it's at <bbox1>,…, at 18s it's at <bbox2> , …, at 22s, it's at <bbox3> Video QA Temporal Video Grounding 4s 7s 16.2s 18s 22s 37s bbox1 bbox2 bbox3 Where and when does a man takes a motorbike away? What are the man at 16.2s <bbox1> doing and when is he stop? Where is he in the video? He rides the motor bike away and stops at 22s. at 16.2s it's at <bbox1>, …, at 18s it's at <bbox2> , …, at 22s, it's at <bbox3>

For agents

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

get_harvested_code_for_paper("2501.08282")
get_code_for_paper("2501.08282")
have("2501.08282")

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

curl https://syntology.ai/api/ran/2501.08282.json

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

[![Syntology run record](https://syntology.ai/api/ran/2501.08282.svg)](https://syntology.ai/paper/2501.08282)

Connect an agent — have() is free.