SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2406.16860 · NeurIPS · 2024

Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs

Sanghyun Woo, Shusheng Yang, Penghao Wu, Ellis Brown, Shengbang Tong, Manoj Middepogu, Charitha Sai, Jihan Akula, Adithya Yang, Xichen Iyer, Ziteng Pan, Rob Wang, and 5 more

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 5 failures of our instrument; 1 no recorded run); 1 of them was also run on the host, not in the jail, by the peer-agreement lane that set its level, 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/2406.16860.json · badge: /api/ran/2406.16860.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
cambrian-mllm/cambrian canonical (classifier label)
classifier basis: classifier_2025
10 ran · 1 no recorded run
FunctionStatusWhere it lives
window_unpartition Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cambrian-mllm/cambrian/cambrian/model/multimodal_encoder/sam/encoder.py
code served (permissive licence) · get_code("7bca1adaded75dc9")
How it ran

Ran in Syntology's jail on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 4, dtype family float, shape (2, 6, 8, 8) as a later run recorded it. The image it ran in was not recorded. The comparison behind this sample's level is a different run: the peer-agreement lane ran it again on the host, not in the jail (report dated 2026-09-05T00:34:38Z). 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

torch.manual_seed(0)

# Parameters for window_unpartition
# windows: [B * num_windows, window_size, window_size, C]
# window_size: int
# pad_hw: (Hp, Wp) - padded height and width
# hw: (H, W) - original height and width

window_size = 4
C = 8  # channels

# Original dimensions
H, W = 6, 8
# Padded dimensions (must be multiples of window_size)
Hp, Wp = 8, 8

# Number of windows in each dimension
num_windows_h = Hp // window_size  # 2
num_windows_w = Wp // window_size  # 2
num_windows = num_windows_h * num_windows_w  # 4

# Batch size
B = 2

# Total number of window patches: B * num_windows
total_windows = B * num_windows  # 2 * 4 = 8

# Create windows tensor: [B * num_windows, window_size, window_size, C]
windows = torch.randn(total_windows, window_size, window_size, C)

ARGS = {
    "windows": windows,
    "window_size": window_size,
    "pad_hw": (Hp, Wp),
    "hw": (H, W),
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 4, dtype family float, shape (2, 6, 8, 8) — shape (2, 6, 8, 8) 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
AggregationBlock Ran
ran on a synthesized input · constructed an object; no result was computed
cambrian-mllm/cambrian/cambrian/model/vision_sampler.py
code served (permissive licence) · get_code("9b244ed4de98b8b3")
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, AggregationBlock: 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
from torch import nn

torch.manual_seed(0)

# Define the dimensions
q_dim = 4
kv_dim = 5
hidden_dim = 8
num_heads = 2
attention_bias = False

# Create the attention module (CrossAttention)
attention_module = nn.Linear(1, 1)  # Dummy module

# Build the ARGS dictionary
ARGS = {
    "attention": True,
    "q_dim": q_dim,
    "kv_dim": kv_dim,
    "hidden_dim": hidden_dim,
    "num_heads": num_heads,
    "attention_bias": attention_bias,
}

# Create an instance to test
block = nn.Module.__new__(type('AggregationBlock', (nn.Module,), {}))

# Actually, let's construct it properly for testing
from torch import nn

class CrossAttention(nn.Module):
    def __init__(self, q_dim, kv_dim, hidden_dim, num_heads, attention_bias=False):
        super().__init__()
        self.hidden_dim = hidden_dim
        self.num_heads = num_heads
        self.head_dim = self.hidden_dim // self.num_heads
        if self.head_dim * self.num_heads != self.hidden_dim:
            raise ValueError(f'hidden_dim must be divisible by num_heads (got `hidden_dim`: {self.hidden_dim} and `num_heads`: {self.num_heads}).')
        self.q_proj = nn.Sequential(nn.LayerNorm(q_dim), nn.Linear(q_dim, self.num_heads * self.head_dim, bias=attention_bias))
        self.k_proj = nn.Sequential(nn.LayerNorm(kv_dim), nn.Linear(kv_dim, self.num_heads * self.head_dim, bias=attention_bias))
        self.v_proj = nn.Sequential(nn.LayerNorm(kv_dim), nn.Linear(kv_dim, self.num_heads * self.head_dim, bias=attention_bias))
        self.o_proj = nn.Linear(self.num_heads * self.head_dim, q_dim, bias=attention_bias)

class MLP(nn.Module):
    def __init__(self, d_in, d_hidden, d_out):
        super().__init__()
        self.linear_1 = nn.Linear(d_in, d_hidden, bias=False)
        self.act = nn.GELU()
        self.linear_2 = nn.Linear(d_hidden, d_out, bias=False)

ARGS = {
    "attention": True,
    "q_dim": 4,
    "kv_dim": 5,
    "hidden_dim": 8,
    "num_heads": 2,
    "attention_bias": 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 AggregationBlock — 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 AggregationBlock 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
CrossAttention Ran
ran on a synthesized input · constructed an object; no result was computed
cambrian-mllm/cambrian/cambrian/model/vision_sampler.py
code served (permissive licence) · get_code("7f1838302ff8b2bc")
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, CrossAttention: 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
from torch import nn

torch.manual_seed(0)

class CrossAttention(nn.Module):

    def __init__(self, q_dim, kv_dim, hidden_dim, num_heads, attention_bias=False):
        super().__init__()
        self.hidden_dim = hidden_dim
        self.num_heads = num_heads
        self.head_dim = self.hidden_dim // self.num_heads
        if self.head_dim * self.num_heads != self.hidden_dim:
            raise ValueError(f'hidden_dim must be divisible by num_heads (got `hidden_dim`: {self.hidden_dim} and `num_heads`: {self.num_heads}).')
        self.q_proj = nn.Sequential(nn.LayerNorm(q_dim), nn.Linear(q_dim, self.num_heads * self.head_dim, bias=attention_bias))
        self.k_proj = nn.Sequential(nn.LayerNorm(kv_dim), nn.Linear(kv_dim, self.num_heads * self.head_dim, bias=attention_bias))
        self.v_proj = nn.Sequential(nn.LayerNorm(kv_dim), nn.Linear(kv_dim, self.num_heads * self.head_dim, bias=attention_bias))
        self.o_proj = nn.Linear(self.num_heads * self.head_dim, q_dim, bias=attention_bias)

    def forward(self, vision_latents, queries, attention_mask):
        bsz, q_len, _ = queries.size()
        bsz, v_len, _ = vision_latents.size()
        query_states = self.q_proj(queries)
        key_states = self.k_proj(vision_latents)
        value_states = self.v_proj(vision_latents)
        query_states = query_states.view(bsz, q_len, self.num_heads, self.head_dim).transpose(1, 2)
        key_states = key_states.view(bsz, v_len, self.num_heads, self.head_dim).transpose(1, 2)
        value_states = value_states.view(bsz, v_len, self.num_heads, self.head_dim).transpose(1, 2)
        if attention_mask is not None:
            if attention_mask.size() != (bsz, 1, q_len, v_len):
                raise ValueError(f'Attention mask should be of size {(bsz, 1, q_len, v_len)}, but is {attention_mask.size()}')
        if query_states.device.type == 'cuda' and attention_mask is not None:
            query_states = query_states.contiguous()
            key_states = key_states.contiguous()
            value_states = value_states.contiguous()
        attn_output = torch.nn.functional.scaled_dot_product_attention(query_states, key_states, value_states, attn_mask=attention_mask)
        attn_output = attn_output.transpose(1, 2).contiguous()
        attn_output = attn_output.reshape(bsz, q_len, self.hidden_dim)
        attn_output = self.o_proj(attn_output)
        return attn_output

q_dim = 4
kv_dim = 5
hidden_dim = 8
num_heads = 2
attention_bias = False

cross_attention = CrossAttention(q_dim, kv_dim, hidden_dim, num_heads, attention_bias)

bsz = 2
q_len = 3
v_len = 4

vision_latents = torch.randn(bsz, v_len, kv_dim)
queries = torch.randn(bsz, q_len, q_dim)
attention_mask = torch.ones(bsz, 1, q_len, v_len)

ARGS = {
    "q_dim": q_dim,
    "kv_dim": kv_dim,
    "hidden_dim": hidden_dim,
    "num_heads": num_heads,
    "attention_bias": attention_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 CrossAttention — 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 CrossAttention 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
MLP Ran
ran on a synthesized input · constructed an object; no result was computed
cambrian-mllm/cambrian/cambrian/model/vision_sampler.py
code served (permissive licence) · get_code("a65d55cf801fa5fb")
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, MLP: 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)

ARGS = {
    "d_in": 4,
    "d_hidden": 3,
    "d_out": 2,
}
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 MLP — 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 MLP 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_chunk Ran
ran, honoured the contract we drafted
cambrian-mllm/cambrian/eval/eval/omni/omni_eval.py
code served (permissive licence) · get_code("0970b04a87d6849f")
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) and returned list, ndim 1, dtype family int; 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 math

def split_list(lst, n):
    """Split a list into n (roughly) equal-sized chunks"""
    chunk_size = math.ceil(lst / n)
    return [[i, i + chunk_size - 1] for i in range(0, lst, chunk_size)]

def get_chunk(lst, n, k):
    chunks = split_list(lst, n)
    return chunks[k]

# Setup code to construct valid arguments
lst_value = 100
n_value = 5
k_value = 2

ARGS = {
    "lst": lst_value,
    "n": n_value,
    "k": k_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 int — 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_rel_pos Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cambrian-mllm/cambrian/cambrian/model/multimodal_encoder/sam/encoder.py
code served (permissive licence) · get_code("733d7f0bedcb74c2")
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) and returned Tensor, ndim 3, dtype family float, shape (3, 4, 4) 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.functional as F

torch.manual_seed(0)

# Create a small relative position embedding tensor
# Shape: (L, C) where L is the sequence length and C is the channel dimension
# Using a small size that will be resized by the function
rel_pos = torch.randn(5, 4)  # 5 positions, 4 channels

# Define query and key sizes
q_size = 3
k_size = 4

ARGS = {
    "q_size": q_size,
    "k_size": k_size,
    "rel_pos": rel_pos,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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, shape (3, 4, 4) — shape (3, 4, 4) 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
split_list Ran
ran, honoured the contract we drafted
cambrian-mllm/cambrian/eval/eval/omni/omni_eval.py
code served (permissive licence) · get_code("9874efcce8755c53")
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) and returned list, ndim 2, dtype family int; 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 math

def split_list(lst, n):
    """Split a list into n (roughly) equal-sized chunks"""
    chunk_size = math.ceil(lst / n)
    return [[i, i + chunk_size - 1] for i in range(0, lst, chunk_size)]

# Construct valid arguments
ARGS = {
    "lst": 100,
    "n": 3
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 2, dtype family int — 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
unmask_attention_mask Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cambrian-mllm/cambrian/cambrian/model/cambrian_arch.py
code served (permissive licence) · get_code("fd89ea82cae32f5d")
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) and returned Tensor, ndim 4, dtype family float, shape (2, 8, 12, 1) 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

torch.manual_seed(0)

# Create a small attention mask tensor
# Shape: (batch_size, height, width, channels) or (batch_size, height, width)
# The function accesses mask.shape[1:3] to get cur_h and cur_w
mask = torch.ones(2, 8, 12, 1)  # batch=2, height=8, width=12, channels=1

# original_size is a tuple of (width, height)
original_size = (6, 4)  # original_w=6, original_h=4

ARGS = {
    "mask": mask,
    "original_size": original_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 4, dtype family float, shape (2, 8, 12, 1) — shape (2, 8, 12, 1) 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
unpad_image Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cambrian-mllm/cambrian/cambrian/model/cambrian_arch.py
code served (permissive licence) · get_code("83d2977826140fdc")
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) and returned Tensor, ndim 3, dtype family float, shape (3, 8, 6) 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

torch.manual_seed(0)

# Create a padded image tensor in CxHxW format
# C=3 (RGB), H=8 (current height), W=6 (current width)
tensor = torch.randn(3, 8, 6)

# Original size as (width, height)
# Using different dimensions to be distinct: width=4, height=5
original_size = (4, 5)

ARGS = {
    "tensor": tensor,
    "original_size": original_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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, shape (3, 8, 6) — shape (3, 8, 6) 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
window_partition Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
cambrian-mllm/cambrian/cambrian/model/multimodal_encoder/sam/encoder.py
code served (permissive licence) · get_code("5693c99bfc32e135")
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) and returned tuple. 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.functional as F

torch.manual_seed(0)

# Create a small input tensor with shape [B, H, W, C]
# B=2 (batch), H=3 (height), W=4 (width), C=5 (channels)
x = torch.randn(2, 3, 4, 5)

# Window size
window_size = 2

ARGS = {
    "x": x,
    "window_size": window_size,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is 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
VisionAggregationLayer Not yet run
unverified (harvested, no recorded run)
cambrian-mllm/cambrian/cambrian/model/vision_sampler.py
code served (permissive licence) · get_code("b25d6d325c57c2a8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 6 are also present in other papers whose code we hold (4 with at least one paper dated earlier, 4 the same month, 6 later) and 5 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
get_chunk 189 other papers
58 dated earlier, 21 the same month, 104 later, 6 same year, order not determinable
2112.05682 (2021-12)
Self-attention Does Not Need $O(n^2)$ Memory
2306.17107 (2023-06)
LLaVAR: Enhanced Visual Instruction Tuning for Text-Rich Image Understanding
2307.10490 (2023-07)
Abusing Images and Sounds for Indirect Instruction Injection in Multi-Modal LLMs
and 186 more not named here
get_rel_pos 47 other papers
33 dated earlier, 1 the same month, 11 later, 2 same year, order not determinable · 5 only as an assembly brick
2304.02643 (2023-04)
Segment Anything
2304.06790 (2023-04)
Inpaint Anything: Segment Anything Meets Image Inpainting
2304.12306 (2023-04)
Segment Anything in Medical Images
and 44 more not named here
window_partition 47 other papers
32 dated earlier, 1 the same month, 12 later, 2 same year, order not determinable · 1 only as an assembly brick
2304.02643 (2023-04)
Segment Anything
2304.06790 (2023-04)
Inpaint Anything: Segment Anything Meets Image Inpainting
2304.12306 (2023-04)
Segment Anything in Medical Images
and 44 more not named here
window_unpartition 46 other papers
32 dated earlier, 1 the same month, 11 later, 2 same year, order not determinable · 1 only as an assembly brick
2304.02643 (2023-04)
Segment Anything
2304.06790 (2023-04)
Inpaint Anything: Segment Anything Meets Image Inpainting
2304.12306 (2023-04)
Segment Anything in Medical Images
and 43 more not named here
unmask_attention_mask 1 other paper
1 later
2410.17434 (2024-10)
LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding
unpad_image 1 other paper
1 later
2410.17434 (2024-10)
LongVU: Spatiotemporal Adaptive Compression for Long Video-Language Understanding

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

We introduce Cambrian-1, a family of multimodal LLMs (MLLMs) designed with a visioncentric approach. While stronger language models can enhance multimodal capabilities, the design choices for vision components are often insufficiently explored and disconnected from visual representation learning research. This gap hinders accurate sensory grounding in realworld scenarios. Our study uses LLMs and visual instruction tuning as an interface to evaluate various visual representations, offering new insights into different models and architecturesself-supervised, strongly supervised, or combinations thereof-based on experiments with over 20 vision encoders. We critically examine existing MLLM benchmarks, address the difficulties involved in consolidating and interpreting results from various tasks, and introduce a new vision-centric benchmark, CV-Bench. To further improve visual grounding, we propose the Spatial Vision Aggregator (SVA), a dynamic and spatially-aware connector that integrates high-resolution vision features with LLMs while reducing the number of tokens. Additionally, we discuss the curation of high-quality visual instruction-tuning data from publicly available sources, emphasizing the importance of data source balancing and distribution ratio. Collectively, Cambrian-1 not only achieves state-of-the-art performance but also serves as a comprehensive, open cookbook for instruction-tuned MLLMs. We provide model weights, code, supporting tools, datasets, and detailed instruction-tuning and evaluation recipes. We hope our release will inspire and accelerate advancements in multimodal systems and visual representation learning.

For agents

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

get_harvested_code_for_paper("2406.16860")
get_code_for_paper("2406.16860")
have("2406.16860")

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

curl https://syntology.ai/api/ran/2406.16860.json

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

[![Syntology run record](https://syntology.ai/api/ran/2406.16860.svg)](https://syntology.ai/paper/2406.16860)

Connect an agent — have() is free.