SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2404.01331 · 2024

LLaVA-Gemma: Accelerating Multimodal Foundation Models with a Compact Language Model

arXiv · PDF · Open in the Atlas

Code that ran

5 of 7 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 0 have no contract verdict; 5 failures of our instrument; 2 no recorded run), measured 2026-11-21; 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/2404.01331.json · badge: /api/ran/2404.01331.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
intellabs/multimodal_cognitive_ai canonical (classifier label)
classifier basis: census
5 ran · 2 no recorded run
FunctionStatusWhere it lives
build_vision_projector Ran
ran; our contract draft was wrong, not the code
intellabs/multimodal_cognitive_ai/LLaVA-Gemma/llava/model/multimodal_projector/builder.py
code served (permissive licence) · get_code("ef4db0da176c674c")
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 Linear. 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 Config:
    def __init__(self, mm_projector_type, mm_hidden_size, hidden_size):
        self.mm_projector_type = mm_projector_type
        self.mm_hidden_size = mm_hidden_size
        self.hidden_size = hidden_size

# Test with 'linear' projector type
config_linear = Config(mm_projector_type='linear', mm_hidden_size=4, hidden_size=3)

ARGS = {
    'config': config_linear,
    '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 Linear — a Linear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Linear 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_ViT_B_16_laion2b_s34b_b88k_property_head_layer_labels Ran
ran; our contract draft was wrong, not the code
intellabs/multimodal_cognitive_ai/CLIP-InterpreT/src/image_property_labels.py
code served (permissive licence) · get_code("318d920a905b1d92")
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 torch
import numpy as np

torch.manual_seed(0)

# Analyzing the function:
# - It takes a single parameter 'property' which is a string
# - It converts it to lowercase
# - It looks it up in a hardcoded properties dict
# - Valid keys are: 'animals', 'locations', 'art', 'subject', 'nature'
# - The function returns a list of layer-head label strings

ARGS = {
    'property': 'animals'
}
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 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
get_ViT_B_16_openai_property_head_layer_labels Ran
ran; our contract draft was wrong, not the code
intellabs/multimodal_cognitive_ai/CLIP-InterpreT/src/image_property_labels.py
code served (permissive licence) · get_code("ddbce3e97829e60c")
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 torch
import torch.nn as nn

torch.manual_seed(0)

# Analyze the function:
# - It takes a single parameter 'property' which is a string
# - It converts it to lowercase
# - It looks up the string in a properties dict
# - Returns a list of layer labels for that property
# Valid properties are 'animals' and 'locations'

ARGS = {
    'property': 'animals'
}
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 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
get_ViT_B_32_datacomp_m_s128m_b4k_property_head_layer_labels Ran
ran; our contract draft was wrong, not the code
intellabs/multimodal_cognitive_ai/CLIP-InterpreT/src/image_property_labels.py
code served (permissive licence) · get_code("c3df7bbbd646a08a")
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 torch
import numpy as np

# Analyze the function: it takes a 'property' parameter (string),
# converts it to lowercase, and returns a list of layer/head labels
# from a properties dictionary based on the property name.

# The function only uses:
# - property.lower() - so property must be a string
# - dictionary lookup with property as key

# Valid property values are 'animals' or 'colors' (case-insensitive)

ARGS = {
    'property': 'animals'
}
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 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
hook_prs_logger Ran
ran; our contract draft was wrong, not the code
intellabs/multimodal_cognitive_ai/CLIP-InterpreT/src/prs_hook.py
code served (permissive licence) · get_code("b9a79e919e4e344c")
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 PRSLogger. 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
import numpy as np

torch.manual_seed(0)

class SimpleAttention(nn.Module):
    def __init__(self, dim=4, num_heads=2):
        super().__init__()
        self.out_proj = nn.Linear(dim, dim)

class SimpleMLP(nn.Module):
    def __init__(self, dim=4):
        super().__init__()
        self.c_proj = nn.Linear(dim, dim)

class SimpleResBlock(nn.Module):
    def __init__(self, dim=4):
        super().__init__()
        self.attn = SimpleAttention(dim)
        self.mlp = SimpleMLP(dim)

class SimpleTransformer(nn.Module):
    def __init__(self, num_layers=2, dim=4):
        super().__init__()
        self.resblocks = nn.ModuleList([SimpleResBlock(dim) for _ in range(num_layers)])

class SimpleVisual(nn.Module):
    def __init__(self, dim=4):
        super().__init__()
        self.transformer = SimpleTransformer(num_layers=2, dim=dim)
        self.ln_post = nn.LayerNorm(dim)
        self.proj = nn.Linear(dim, 3)

class SimpleModel(nn.Module):
    def __init__(self, dim=4):
        super().__init__()
        self.visual = SimpleVisual(dim)

class SimpleHookManager:
    def __init__(self):
        self.hooks = {}
    
    def register(self, pattern, callback):
        self.hooks[pattern] = callback

model = SimpleModel(dim=4)
model.hook_manager = SimpleHookManager()

device = 'cpu'

ARGS = {
    'model': model,
    'device': device,
}
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 PRSLogger — a PRSLogger output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a PRSLogger 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_tower Not yet run
unverified (harvested, no recorded run)
intellabs/multimodal_cognitive_ai/LLaVA-Gemma/llava/model/multimodal_encoder/builder.py
code served (permissive licence) · get_code("4a5d89f6478c0361")
How it ran

no invocation recorded.

image_grid Not yet run
unverified (harvested, no recorded run)
intellabs/multimodal_cognitive_ai/CLIP-InterpreT/src/helper.py
code served (permissive licence) · get_code("d5db2b888b479e94")
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, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 later) and 6 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
build_vision_projector 17 other papers
3 dated earlier, 14 later · 13 only as an assembly brick
2301.12597 (2023-01)
BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models
2304.08485 (2023-04)
Visual Instruction Tuning
2402.18695 (2024-02)
Grounding Language Models for Visual Entity Recognition
and 14 more not named here

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 2026-11-21.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We train a suite of multimodal foundation models (MMFM) using the popular LLaVA framework with the recently released Gemma family of large language models (LLMs). Of particular interest is the 2B parameter Gemma model, which provides opportunities to construct capable small-scale MMFMs. In line with findings from other papers in this space, we test the effect of ablating three design features: pretraining the connector, utilizing a more powerful image backbone, and increasing the size of the language backbone. The resulting models, which we call LLaVA-Gemma, exhibit moderate performance on an array of evaluations, but fail to improve past the current comparably sized SOTA models. Closer analysis of performance shows mixed effects; skipping pretraining tends to reduce performance, larger vision models sometimes improve performance, and increasing language model size has inconsistent effects. We publicly release training recipes, code and weights for our models for the LLaVA-Gemma models.

For agents

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

get_harvested_code_for_paper("2404.01331")
get_code_for_paper("2404.01331")
have("2404.01331")

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

curl https://syntology.ai/api/ran/2404.01331.json

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

[![Syntology run record](https://syntology.ai/api/ran/2404.01331.svg)](https://syntology.ai/paper/2404.01331)

Connect an agent — have() is free.