SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2103.03206 · ICML · 2021

Perceiver: General Perception with Iterative Attention

Andrew Zisserman, Oriol Vinyals, Andrew Brock, Andrew Jaegle, Felix Gimeno, Joao Carreira

arXiv · PDF · Open in the Atlas

Code that ran

41 of 55 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (20 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 2 violated it and 28 have no contract verdict; 11 failures of our instrument; 14 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; 3 contract verdicts on these samples were measured on the host before the contract probe was jailed (2026-09-18T14:29:13Z) or are undated; 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/2103.03206.json · badge: /api/ran/2103.03206.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
lucidrains/perceiver-pytorch reimplementation (classifier label)
classifier basis: classifier_2025
8 ran · 3 no recorded run
Rishit-dagli/Perceiver reimplementation (classifier label)
classifier basis: classifier_2025
5 ran · 1 no recorded run
kietngt00/hmdb51-recognition community
Papers with Code archive
5 ran · 0 no recorded run
fac2003/perceiver-multi-modality-pytorch community
Papers with Code archive
5 ran · 0 no recorded run
louislva/deepmind-perceiver reimplementation (classifier label)
classifier basis: classifier_2025
5 ran · 0 no recorded run
google-research/perceiver-ar extension (classifier label)
classifier basis: classifier_2025
4 ran · 0 no recorded run
sooheon/perceiver-jax community
Papers with Code archive
3 ran · 2 no recorded run
clementpoiret/Perceiver_MNIST community
Papers with Code archive
2 ran · 3 no recorded run
towhee-io/towhee community
Papers with Code archive
2 ran · 0 no recorded run
Xianchao-Wu/perceiver-pytorch reimplementation (classifier label)
classifier basis: classifier_2025
1 ran · 4 no recorded run
copy not recorded community
not in the paper's declared repositories
1 ran · 0 no recorded run
deepmind/deepmind-research marked official in Papers with Code
Papers with Code archive
0 ran · 1 no recorded run
FunctionStatusWhere it lives
default Ran
ran, violated the contract we drafted
this paper's copy was not recorded; identical code first harvested from stanliu96/RATD
pointer only · get_code("7c49b92babb27a7d")
How it ran

The body of this sample is withheld: this paper's copy of the code was not recorded (its link names no repository or source path), so no licence was read for it. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Attention Ran
ran on a synthesized input · constructed an object; no result was computed
Rishit-dagli/Perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("2b6160b81f7900b0")
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, Attention: 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 tensorflow as tf
import numpy as np
from einops import rearrange, repeat

# Set seed for reproducibility
np.random.seed(0)
tf.random.set_seed(0)

# Create small test tensors
batch_size = 2
seq_len = 3
query_dim = 16
context_dim = 16
heads = 4
dim_head = 4
dropout = 0.1

# Create dummy input tensor: (batch, seq_len, query_dim)
x = tf.constant(np.random.randn(batch_size, seq_len, query_dim).astype(np.float32))

# Create dummy context tensor: (batch, seq_len, context_dim)
context = tf.constant(np.random.randn(batch_size, seq_len, context_dim).astype(np.float32))

# Create dummy mask tensor: (batch, seq_len)
mask = tf.constant(np.ones((batch_size, seq_len), dtype=bool))

ARGS = {
    'query_dim': query_dim,
    'context_dim': context_dim,
    'heads': heads,
    'dim_head': dim_head,
    'dropout': dropout,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Attention — 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 Attention 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
Attention Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("1f325569b6d0c2df")
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, Attention: 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
from torch import nn

torch.manual_seed(0)

# Build the Attention module for testing
attention_module = nn.Module()

# Define small distinct tensor dimensions for x and context
batch_size = 2
seq_len_x = 3
seq_len_context = 4
query_dim = 8
context_dim = 16
heads = 4
dim_head = 2
dropout = 0.1

# Create input tensors
x = torch.randn(batch_size, seq_len_x, query_dim)
context = torch.randn(batch_size, seq_len_context, context_dim)

# Create attention mask (batch_size, seq_len_context)
mask = torch.ones(batch_size, seq_len_context, dtype=torch.bool)
mask[0, -1] = False  # Mask out one position

ARGS = {
    'query_dim': query_dim,
    'context_dim': context_dim,
    'heads': heads,
    'dim_head': dim_head,
    'dropout': dropout,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Attention — 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 Attention 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
Attention Ran
ran on a synthesized input
kietngt00/hmdb51-recognition/model/attention.py
pointer only (licence: NONE) · get_code("407c9004f3d03831")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, kietngt00/hmdb51-recognition when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Attention Ran
ran on a synthesized input · constructed an object; no result was computed
sooheon/perceiver-jax/perceiver_jax/perceiver_jax.py
code served (permissive licence) · get_code("9a7903b6b454a21a")
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, Attention: 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 jax
import jax.numpy as jnp
from flax import linen as nn

# Set random seed for reproducibility
key = jax.random.PRNGKey(0)

# Create tiny input tensor: (batch=2, sequence_length=3, feature_dim=512)
# where feature_dim = heads * head_features = 8 * 64
x = jnp.ones((2, 3, 512))

# Optional context tensor (same shape as x for this test)
context = jnp.ones((2, 3, 512))

# Optional mask tensor: (batch=2, 1, seq_len=3, seq_len=3)
mask = jnp.ones((2, 1, 3, 3))

ARGS = {
    "heads": 8,
    "head_features": 64,
    "dropout": 0.1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, flax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Attention — 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 Attention 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
Attention Ran
ran on a synthesized input · constructed an object; no result was computed
fac2003/perceiver-multi-modality-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("b908a940626ba009")
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, Attention: 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
from functools import wraps, partial
from einops import rearrange, repeat
from torch.utils.checkpoint import checkpoint

torch.manual_seed(0)

def exists(val):
    return val is not None

def default(val, d):
    return val if exists(val) else d

class Attention(nn.Module):

    def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.0):
        super().__init__()
        inner_dim = dim_head * heads
        context_dim = default(context_dim, query_dim)
        self.scale = dim_head ** (-0.5)
        self.heads = heads
        self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
        self.to_kv = nn.Linear(context_dim, inner_dim * 2, bias=False)
        self.to_out = nn.Sequential(nn.Linear(inner_dim, query_dim), nn.Dropout(dropout))

    def forward(self, x, context=None, mask=None):
        h = self.heads
        q = self.to_q(x)
        context = default(context, x)
        k, v = self.to_kv(context).chunk(2, dim=-1)
        k = k.float()
        q = q.float()
        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
        sim = checkpoint(partial(torch.einsum, 'b i d, b j d -> b i j'), q, k) * self.scale
        if exists(mask):
            mask = rearrange(mask, 'b ... -> b (...)')
            max_neg_value = -torch.finfo(sim.dtype).max
            mask = repeat(mask, 'b j -> (b h) () j', h=h)
            sim.masked_fill_(~mask, max_neg_value)
        attn = sim.softmax(dim=-1)
        out = checkpoint(partial(torch.einsum, 'b i j, b j d -> b i d'), attn, v)
        out = out.type(x.dtype)
        out = rearrange(out, '(b h) n d -> b n (h d)', h=h)
        return self.to_out(out)

# Create minimal valid arguments
query_dim = 16
context_dim = 16
heads = 4
dim_head = 4
dropout = 0.1

# Create tensors for forward pass
batch_size = 2
seq_len_q = 3
seq_len_kv = 5

x = torch.randn(batch_size, seq_len_q, query_dim)
context = torch.randn(batch_size, seq_len_kv, context_dim)
mask = torch.ones(batch_size, seq_len_kv, dtype=torch.bool)

ARGS = {
    'query_dim': query_dim,
    'context_dim': context_dim,
    'heads': heads,
    'dim_head': dim_head,
    'dropout': dropout,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Attention — 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 Attention 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
Attention Ran
ran on a synthesized input
Xianchao-Wu/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("dab3a202eec902f5")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 5, 32) 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
from torch import nn, einsum
from einops import rearrange, repeat

def default(val, d):
    return val if exists(val) else d

def exists(val):
    return val is not None

class Attention(nn.Module):

    def __init__(self, query_dim, context_dim=None, heads=8, dim_head=64, dropout=0.0):
        super().__init__()
        inner_dim = dim_head * heads
        context_dim = default(context_dim, query_dim)
        self.scale = dim_head ** (-0.5)
        self.heads = heads
        self.to_q = nn.Linear(query_dim, inner_dim, bias=False)
        self.to_kv = nn.Linear(context_dim, inner_dim * 2, bias=False)
        self.to_out = nn.Sequential(nn.Linear(inner_dim, query_dim), nn.Dropout(dropout))

    def forward(self, x, context=None, mask=None):
        h = self.heads
        q = self.to_q(x)
        context = default(context, x)
        k, v = self.to_kv(context).chunk(2, dim=-1)
        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> (b h) n d', h=h), (q, k, v))
        sim = einsum('b i d, b j d -> b i j', q, k) * self.scale
        if exists(mask):
            mask = rearrange(mask, 'b ... -> b (...)')
            max_neg_value = -torch.finfo(sim.dtype).max
            mask = repeat(mask, 'b j -> (b h) () j', h=h)
            sim.masked_fill_(~mask, max_neg_value)
        attn = sim.softmax(dim=-1)
        out = einsum('b i j, b j d -> b i d', attn, v)
        out = rearrange(out, '(b h) n d -> b n (h d)', h=h)
        return self.to_out(out)


INIT_ARGS = {
    'query_dim': 32,
    'context_dim': 32,
    'heads': 4,
    'dim_head': 8,
    'dropout': 0.1,
}

ARGS = {
    'x': torch.randn(2, 5, 32),
    'context': torch.randn(2, 7, 32),
    'mask': torch.ones(2, 7, dtype=torch.bool),
}

if __name__ == '__main__':
    attention = Attention(**INIT_ARGS)
    output = attention(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 5, 32)")
    assert output.shape == (2, 5, 32), f"Shape mismatch: {output.shape}"
    print("Success!")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
einops, 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 (2, 5, 32) — shape (2, 5, 32) 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
CrossAttention Ran
ran on a synthesized input · constructed an object; no result was computed
kietngt00/hmdb51-recognition/model/attention.py
pointer only (licence: NONE) · get_code("122a2bb1f3342e5c")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, kietngt00/hmdb51-recognition when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

CrossAttention Ran
ran on a synthesized input
towhee-io/towhee/towhee/models/perceiver/cross_attention.py
code served (permissive licence) · get_code("ce5db0a13e19e003")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 3, 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
from torch import nn

class MultiHeadAttention(nn.Module):
    def __init__(self, num_q_channels: int, num_kv_channels: int, num_heads: int, dropout: float):
        super().__init__()
        self.attention = nn.MultiheadAttention(embed_dim=num_q_channels, num_heads=num_heads, kdim=num_kv_channels, vdim=num_kv_channels, dropout=dropout, batch_first=True)

    def forward(self, x_q, x_kv, pad_mask=None, attn_mask=None):
        return self.attention(x_q, x_kv, x_kv, key_padding_mask=pad_mask, attn_mask=attn_mask)[0]

class CrossAttention(nn.Module):
    def __init__(self, num_q_channels: int, num_kv_channels: int, num_heads: int, dropout: float):
        super().__init__()
        self.q_norm = nn.LayerNorm(num_q_channels)
        self.kv_norm = nn.LayerNorm(num_kv_channels)
        self.attention = MultiHeadAttention(num_q_channels=num_q_channels, num_kv_channels=num_kv_channels, num_heads=num_heads, dropout=dropout)

    def forward(self, x_q, x_kv, pad_mask=None, attn_mask=None):
        x_q = self.q_norm(x_q)
        x_kv = self.kv_norm(x_kv)
        return self.attention(x_q, x_kv, pad_mask=pad_mask, attn_mask=attn_mask)

INIT_ARGS = {
    "num_q_channels": 4,
    "num_kv_channels": 6,
    "num_heads": 2,
    "dropout": 0.1,
}

ARGS = {
    "x_q": torch.randn(2, 3, 4),
    "x_kv": torch.randn(2, 5, 6),
    "pad_mask": None,
    "attn_mask": None,
}

if __name__ == "__main__":
    model = CrossAttention(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shape:", output.shape)
    print("Test passed!")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 (2, 3, 4) — shape (2, 3, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
FeedForward Ran
ran on a synthesized input · constructed an object; no result was computed
Rishit-dagli/Perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("ef7ed09adef36f64")
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, FeedForward: 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 tensorflow as tf
import numpy as np

# Set random seed for reproducibility
tf.random.set_seed(0)
np.random.seed(0)

# Define the GEGLU and FeedForward classes as provided
class GEGLU(tf.keras.layers.Layer):
    def call(self, x):
        x, gates = tf.split(x, 2, axis=-1)
        return x * tf.nn.gelu(gates)

class FeedForward(tf.keras.layers.Layer):
    def __init__(self, dim, mult=4, dropout=0.0):
        super(FeedForward, self).__init__()
        self.net = tf.keras.Sequential([
            tf.keras.layers.Dense(dim * mult * 2, input_dim=dim),
            GEGLU(),
            tf.keras.layers.Dropout(dropout),
            tf.keras.layers.Dense(dim, input_dim=dim * mult)
        ])

    def call(self, inputs):
        return self.net(inputs)

# Build ARGS dictionary with parameters for FeedForward
ARGS = {
    "dim": 4,
    "mult": 4,
    "dropout": 0.1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type FeedForward — 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 FeedForward 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
FeedForward Ran
ran on a synthesized input · constructed an object; no result was computed
clementpoiret/Perceiver_MNIST/utils/perceiver.py
pointer only (licence: NONE) · get_code("d7a83e087ca450f4")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, clementpoiret/Perceiver_MNIST when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

FeedForward Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("9cc277014f5865d5")
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, FeedForward: 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
from torch import nn
import torch.nn.functional as F

torch.manual_seed(0)

ARGS = {
    'dim': 16,
    'mult': 4,
    'dropout': 0.1
}
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 FeedForward — 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 FeedForward 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
FeedForward Ran
ran on a synthesized input · constructed an object; no result was computed
kietngt00/hmdb51-recognition/model/attention.py
pointer only (licence: NONE) · get_code("b04a1cb2350af23e")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, kietngt00/hmdb51-recognition when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

FeedForward Ran
ran on a synthesized input · constructed an object; no result was computed
sooheon/perceiver-jax/perceiver_jax/perceiver_jax.py
code served (permissive licence) · get_code("829093ff624cc3ea")
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, FeedForward: 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 jax
import jax.numpy as jnp
from flax import linen as nn

# Define the FeedForward module
class FeedForward(nn.Module):
    mult: int = 4
    dropout: float = 0.0

    @nn.compact
    def __call__(self, x, deterministic=False):
        features = x.shape[-1]
        x = nn.Dense(features * self.mult)(x)
        x = nn.gelu(x)
        x = nn.Dropout(self.dropout)(x, deterministic=deterministic)
        x = nn.Dense(features)(x)
        return x

# Create a small input tensor (batch_size=2, features=4)
x = jnp.ones((2, 4), dtype=jnp.float32)

# Initialize the module
ff_module = FeedForward(mult=4, dropout=0.1)

# Initialize module parameters
key = jax.random.PRNGKey(0)
params = ff_module.init(key, x, deterministic=False)

ARGS = {
    "mult": 4,
    "dropout": 0.1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
flax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type FeedForward — 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 FeedForward 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
GEGLU Ran
ran on a synthesized input
Rishit-dagli/Perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("7bcff3ae8d60e20e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned EagerTensor, ndim 3, 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 tensorflow as tf
import numpy as np

# GEGLU expects input with double the feature dimension, which it splits in half
# We'll use batch=2, seq_len=3, and the input will have dim=8 (splits into 2x4)

INIT_ARGS = {}

# The call method expects x with shape [..., 2*d] where it splits into two parts of size d
# Using batch=2, seq_len=3, features=8 (will be split into 2x4)
x_input = tf.constant(np.random.randn(2, 3, 8).astype(np.float32))

ARGS = {'x': x_input}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
GEGLU Ran
ran on a synthesized input
clementpoiret/Perceiver_MNIST/utils/perceiver.py
pointer only (licence: NONE) · get_code("7852fe36c3e3b1f9")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, clementpoiret/Perceiver_MNIST when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

GEGLU Ran
ran on a synthesized input
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("7dda3cde0f4a0d98")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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

INIT_ARGS = {}

# GEGLU expects input where last dimension will be split in half
# So if we want output features of size 4, input must be size 8
x = torch.randn(2, 3, 8)

ARGS = {'x': x}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
MultiHeadAttention Ran
ran on a synthesized input
towhee-io/towhee/towhee/models/perceiver/cross_attention.py
code served (permissive licence) · get_code("055997cbf4c57dd0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 5, 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
from torch import nn

class MultiHeadAttention(nn.Module):
    """
    Multi head attention for Perceiver https://arxiv.org/pdf/2103.03206.pdf.
    Args:
        num_q_channels (`int`):
            Number of q channels.
        num_kv_channels (`int`):
            Number of k or v channels. k has the same channels as v.
        num_heads (`int`):
            Number of parallel attention heads.
        dropout (`nn.Module`):
            Dropout probability.
    """

    def __init__(self, num_q_channels: int, num_kv_channels: int, num_heads: int, dropout: float):
        super().__init__()
        self.attention = nn.MultiheadAttention(embed_dim=num_q_channels, num_heads=num_heads, kdim=num_kv_channels, vdim=num_kv_channels, dropout=dropout, batch_first=True)

    def forward(self, x_q, x_kv, pad_mask=None, attn_mask=None):
        """
        Forward function.
        Args:
            x_q (`Tensor`):
                Query embeddings.
            x_kv (`Tensor`):
                Key embeddings. Key equals value.
            pad_mask (`int`):
                Padding mask.
            attn_mask (`nn.Module`):
                Attention mask.
        """
        return self.attention(x_q, x_kv, x_kv, key_padding_mask=pad_mask, attn_mask=attn_mask)[0]


INIT_ARGS = {
    "num_q_channels": 4,
    "num_kv_channels": 6,
    "num_heads": 2,
    "dropout": 0.1,
}

ARGS = {
    "x_q": torch.randn(2, 5, 4),
    "x_kv": torch.randn(2, 7, 6),
    "pad_mask": None,
    "attn_mask": None,
}

if __name__ == "__main__":
    model = MultiHeadAttention(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shape:", output.shape)
    print("Expected shape: (2, 5, 4)")
    print("Test passed!" if output.shape == (2, 5, 4) else "Test failed!")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 (2, 5, 4) — shape (2, 5, 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Perceiver Ran
ran on a synthesized input · constructed an object; no result was computed
louislva/deepmind-perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("765c0c81701efcfc")
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, Perceiver: 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
from torch import nn
import math

torch.manual_seed(0)

ARGS = {
    "input_channels": 3,
    "input_shape": (4, 5),
    "fourier_bands": 2,
    "latents": 6,
    "d_model": 8,
    "heads": 2,
    "latent_blocks": 2,
    "dropout": 0.1,
    "layers": 3,
}
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 Perceiver — 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 Perceiver 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
Perceiver Ran
ran on a synthesized input · constructed an object; no result was computed
Rishit-dagli/Perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("49b3ac32977e5194")
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, Perceiver: 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 tensorflow as tf
import math

ARGS = {
    "num_freq_bands": 4,
    "depth": 2,
    "max_freq": 10,
    "freq_base": 2,
    "input_channels": 3,
    "input_axis": 2,
    "num_latents": 8,
    "latent_dim": 16,
    "cross_heads": 1,
    "latent_heads": 2,
    "cross_dim_head": 8,
    "latent_dim_head": 8,
    "num_classes": 10,
    "attn_dropout": 0.1,
    "ff_dropout": 0.1,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Perceiver — 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 Perceiver 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
Perceiver Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("5a2b853142bcff22")
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, Perceiver: 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 = {
    "num_freq_bands": 4,
    "depth": 2,
    "max_freq": 10.0,
    "input_channels": 3,
    "input_axis": 2,
    "num_latents": 8,
    "latent_dim": 16,
    "cross_heads": 1,
    "latent_heads": 2,
    "cross_dim_head": 8,
    "latent_dim_head": 8,
    "num_classes": 10,
    "attn_dropout": 0.0,
    "ff_dropout": 0.0,
    "weight_tie_layers": False,
    "fourier_encode_data": True,
    "self_per_cross_attn": 1,
    "final_classifier_head": True,
    "inverted_cross_attn": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Perceiver — 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 Perceiver 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
Perceiver Ran
ran on a synthesized input · constructed an object; no result was computed
fac2003/perceiver-multi-modality-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("ac1c3509b3aee38c")
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, Perceiver: 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 = {
    'num_freq_bands': 2,
    'depth': 2,
    'max_freq': 10.0,
    'freq_base': 2,
    'input_channels': 3,
    'input_axis': 2,
    'num_latents': 4,
    'latent_dim': 8,
    'cross_heads': 1,
    'latent_heads': 2,
    'cross_dim_head': 4,
    'latent_dim_head': 4,
    'num_classes': 10,
    'attn_dropout': 0.0,
    'ff_dropout': 0.0,
    'weight_tie_layers': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Perceiver — 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 Perceiver 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
PerceiverAttentionBlock Ran
ran on a synthesized input
louislva/deepmind-perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("28861a4c4ddbb06a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, 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

# INIT_ARGS: parameters for PerceiverAttentionBlock.__init__
INIT_ARGS = {
    'd_model': 8,
    'heads': 2,
    'dropout': 0.1
}

# ARGS: parameters for the forward method
# x: input data (sequence_length_x, batch_size, d_model)
# z_input: latent array (sequence_length_z, batch_size, d_model)
ARGS = {
    'x': torch.randn(5, 2, 8),  # 5 sequence length for x, batch 2, d_model 8
    'z_input': torch.randn(3, 2, 8)  # 3 sequence length for z, batch 2, d_model 8
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, 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 — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
PerceiverBlock Ran
ran on a synthesized input · constructed an object; no result was computed
louislva/deepmind-perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("28e6bb621fe7b050")
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, PerceiverBlock: 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)

# Parameters for PerceiverBlock
d_model = 4
latent_blocks = 2
dropout = 0.1
heads = 2

# Create tensors for forward pass
# x: input tensor, shape (seq_len, batch, d_model)
# z: latent tensor, shape (latent_len, batch, d_model)
batch_size = 2
seq_len = 3
latent_len = 5

x = torch.randn(seq_len, batch_size, d_model)
z = torch.randn(latent_len, batch_size, d_model)

ARGS = {
    "d_model": d_model,
    "latent_blocks": latent_blocks,
    "dropout": dropout,
    "heads": heads,
}
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 PerceiverBlock — 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 PerceiverBlock 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
PerceiverBlockRepeater Ran
ran on a synthesized input · constructed an object; no result was computed
louislva/deepmind-perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("69c8bf0af95ad719")
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, PerceiverBlockRepeater: 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)

# Create a minimal module that takes (x, z) and returns z
class SimplePerceiver(nn.Module):
    def __init__(self):
        super(SimplePerceiver, self).__init__()
        self.linear = nn.Linear(4, 4)
    
    def forward(self, x, z):
        # x shape: (batch, seq_len, dim)
        # z shape: (batch, latent_dim)
        # Return modified z
        return self.linear(z)

# Create the module to be repeated
module = SimplePerceiver()

# Create the ARGS dictionary
ARGS = {
    "module": module,
    "repeats": 3
}
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 PerceiverBlockRepeater — 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 PerceiverBlockRepeater 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
PreNorm Ran
ran on a synthesized input · constructed an object; no result was computed
Rishit-dagli/Perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("05438d49f4565e5e")
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, PreNorm: 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 tensorflow as tf
import numpy as np

# Create a simple callable that acts as the fn parameter
class SimpleFn(tf.keras.layers.Layer):
    def __init__(self):
        super(SimpleFn, self).__init__()
    
    def call(self, x):
        # The fn is called with x in the PreNorm.call method
        return x * 2

# Instantiate the fn
fn = SimpleFn()

# Define ARGS with all parameters
ARGS = {
    'dim': 4,
    'fn': fn,
    'context_dim': None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type PreNorm — 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 PreNorm 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
PreNorm Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("3c2027fdf007f1d3")
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, PreNorm: 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
from torch import nn

torch.manual_seed(0)

class DummyFunction(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.linear = nn.Linear(dim, dim)
    
    def forward(self, x, **kwargs):
        return self.linear(x)

dim = 4
context_dim = 3
fn = DummyFunction(dim)

ARGS = {
    'dim': dim,
    'fn': fn,
    'context_dim': context_dim
}
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 PreNorm — 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 PreNorm 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
PreNorm Ran
ran on a synthesized input · constructed an object; no result was computed
kietngt00/hmdb51-recognition/model/attention.py
pointer only (licence: NONE) · get_code("9a15eee5a72a53ec")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, kietngt00/hmdb51-recognition when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

PreNorm Ran
ran on a synthesized input · constructed an object; no result was computed
fac2003/perceiver-multi-modality-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("6f367de5d5a82e44")
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, PreNorm: 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
from torch import nn

torch.manual_seed(0)

# Define a minimal callable module to use as fn
class SimpleModule(nn.Module):
    def __init__(self, dim):
        super().__init__()
        self.linear = nn.Linear(dim, dim)
    
    def forward(self, x, **kwargs):
        return self.linear(x)

# Create the arguments
dim = 4
fn = SimpleModule(dim)
context_dim = 3

ARGS = {
    'dim': dim,
    'fn': fn,
    'context_dim': context_dim
}
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 PreNorm — 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 PreNorm 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
cache_fn Ran
ran; our contract draft was wrong, not the code
fac2003/perceiver-multi-modality-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("f449e2a5647debf4")
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 function. 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
from functools import wraps, partial

torch.manual_seed(0)

# Define a simple function to be cached
def sample_function(x, y):
    """A simple function that performs some computation."""
    return x + y

# Test the cache_fn decorator
@wraps(sample_function)
def wrapped_sample_function(*args, _cache=True, **kwargs):
    """Wrapped version for testing."""
    cache = None
    
    @wraps(sample_function)
    def cached_fn(*inner_args, _cache=True, **inner_kwargs):
        if not _cache:
            return sample_function(*inner_args, **inner_kwargs)
        nonlocal cache
        if cache is not None:
            return cache
        cache = sample_function(*inner_args, **inner_kwargs)
        return cache
    return cached_fn(*args, _cache=_cache, **kwargs)

# Create test tensors
x = torch.randn(2, 4)
y = torch.randn(2, 4)

# The function cache_fn takes one parameter: f (a function to be decorated)
# We need to pass a callable function
def test_func(a, b):
    """Simple test function that performs computation."""
    return a * b + torch.ones_like(a)

ARGS = {
    'f': test_func
}
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 function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function 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
fill_diagonal Ran
ran; our contract draft was wrong, not the code
google-research/perceiver-ar/perceiver_ar/perceiver_ar_model.py
code served (permissive licence) · get_code("d1297a0c047d6cb7")
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 ArrayImpl, ndim 3, dtype family float, shape (2, 3, 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 jax.numpy as jnp
import jax

# Create a small array with at least 2 dimensions
# Using shape (2, 3, 4) to have distinct sizes for different dimensions
a = jnp.ones((2, 3, 4))

# Create a scalar value to fill the diagonal with
val = 5.0

ARGS = {
    "a": a,
    "val": val,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 3, dtype family float, shape (2, 3, 4) — shape (2, 3, 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
fourier_encode Ran
ran; our contract draft was wrong, not the code
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("7764f517d3e13e73")
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 Tensor, ndim 3, dtype family float, shape (2, 4, 9) 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
from math import pi

torch.manual_seed(0)

# x: a tensor of arbitrary shape that will be rearranged to (..., 1)
# Let's use a simple 2D tensor with batch and feature dimensions
x = torch.randn(2, 4)

# max_freq: a float representing the maximum frequency for Fourier encoding
max_freq = 10.0

# num_bands: an int representing the number of frequency bands
num_bands = 4

ARGS = {
    'x': x,
    'max_freq': max_freq,
    'num_bands': num_bands
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, 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 (2, 4, 9) — shape (2, 4, 9) 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
fourier_encode Ran
ran; our contract draft was wrong, not the code
kietngt00/hmdb51-recognition/model/attention.py
pointer only (licence: NONE) · get_code("a72f0467d5ab04ca")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, kietngt00/hmdb51-recognition when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

fourier_encode Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
sooheon/perceiver-jax/perceiver_jax/perceiver_jax.py
code served (permissive licence) · get_code("8d50746971d74b4a")
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 ArrayImpl, ndim 3, dtype family float, shape (2, 4, 9) 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 jax.numpy as jnp
import numpy as np

# Create a simple input array for fourier_encode
# The function expects a jnp.ndarray of arbitrary shape
# Let's create a small 2D array (batch_size=2, features=4)
x_input = jnp.array([
    [1.0, 2.0, 3.0, 4.0],
    [5.0, 6.0, 7.0, 8.0]
], dtype=jnp.float32)

num_encodings_value = 4

ARGS = {
    "x": x_input,
    "num_encodings": num_encodings_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 3, dtype family float, shape (2, 4, 9) — shape (2, 4, 9) 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
fourier_encode Ran
ran; our contract draft was wrong, not the code
fac2003/perceiver-multi-modality-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("2db56d1fa509f735")
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 Tensor, ndim 3, dtype family float, shape (2, 4, 9) 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
from math import pi, log

torch.manual_seed(0)

# Create a small input tensor
# The function expects x to be a tensor that can be unsqueezed
# Shape: (batch_size, feature_dim) = (2, 4)
x = torch.randn(2, 4)

# max_freq: a scalar value for the maximum frequency
max_freq = 10.0

# num_bands: number of frequency bands
num_bands = 4

# base: base for logspace
base = 2

ARGS = {
    "x": x,
    "max_freq": max_freq,
    "num_bands": num_bands,
    "base": base
}
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 Tensor, ndim 3, dtype family float, shape (2, 4, 9) — shape (2, 4, 9) 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
fourier_features Ran
ran; our contract draft was wrong, not the code
louislva/deepmind-perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("088d7be4e21dbe4f")
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 Tensor, ndim 3, dtype family float, shape (8, 4, 3) 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 math

torch.manual_seed(0)

# Construct valid arguments for fourier_features
# Parameters: shape (tuple of ints), bands (int)

ARGS = {
    "shape": (4, 3),  # 2D spatial dimensions with distinct sizes
    "bands": 2        # number of frequency bands
}
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 Tensor, ndim 3, dtype family float, shape (8, 4, 3) — shape (8, 4, 3) 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
get_sequence_length Ran
ran, violated the contract we drafted
google-research/perceiver-ar/perceiver_ar/perceiver_ar_model.py
code served (permissive licence) · get_code("3fa4cf728bbb0dff")
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 ArrayImpl, ndim 0, dtype family int, 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 jax.numpy as jnp
import numpy as np

# Create a small sequence with some zeros
# The function expects a sequence where zeros mark the end
# Let's create a sequence: [1, 2, 3, 0] - length should be 3
sequence = jnp.array([1, 2, 3, 0], dtype=jnp.float32)

ARGS = {
    "sequence": sequence
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 0, dtype family int, 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
get_sweep Ran
ran; our contract draft was wrong, not the code
google-research/perceiver-ar/perceiver_ar/experiment.py
code served (permissive licence) · get_code("3ce096de610c4729")
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 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

torch.manual_seed(0)

# The function get_sweep takes three parameters:
# - sweep_name: str
# - train_device_count: int
# - eval_device_count: int

# Looking at the function body, it:
# 1. Checks if sweep_name matches certain strings
# 2. Returns a list of tuples with configuration parameters
# 3. Calls _get_batch_sweep with train_batch_sizes, train_batch_sizes_per_device, 
#    train_device_count, eval_batch_size_per_device, eval_device_count

# The function doesn't actually use train_device_count and eval_device_count 
# in the main logic shown, but they are passed to _get_batch_sweep.
# Looking at _get_batch_sweep, it uses train_device_count for division checks.

ARGS = {
    'sweep_name': 'dummy_use_positions_from_data',
    'train_device_count': 8,
    'eval_device_count': 1,
}
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 list, ndim 2, 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
l1norm Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("f1ef7aeee4ed2d74")
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 2, dtype family float, shape (2, 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 tensor for testing
t = torch.randn(2, 4)

# Define ARGS dict with all parameters for l1norm
ARGS = {
    't': t,
    'dim': -1,
    'eps': 1e-08
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
l1norm Ran
ran; our contract draft was wrong, not the code
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("52649c5bdcbe1d0a")
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 Tensor, ndim 2, dtype family float, shape (2, 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 tensor for testing l1norm
t = torch.randn(2, 4)

# Define ARGS with all parameters
ARGS = {
    't': t,
    'dim': -1,
    'eps': 1e-08
}
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 Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 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
truncate_sequence Ran
ran; our contract draft was wrong, not the code
google-research/perceiver-ar/perceiver_ar/perceiver_ar_model.py
code served (permissive licence) · get_code("1357068f980451b5")
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 ArrayImpl, ndim 1, dtype family int, shape (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 jax.numpy as jnp
import numpy as np

# Create a small test sequence
# The function expects a JAX array with integer values
# It finds the first zero to determine sequence length
# Let's create a sequence: [5, 3, 7, 0] - length should be 3
sequence = jnp.array([5, 3, 7, 0], dtype=jnp.int32)

ARGS = {
    "sequence": sequence
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family int, shape (4,) — shape (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
Attention Not yet run
unverified (harvested, no recorded run)
clementpoiret/Perceiver_MNIST/utils/perceiver.py
pointer only (licence: NONE) · get_code("c78772b126a7d1bf")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, clementpoiret/Perceiver_MNIST when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Perceiver Not yet run
unverified (harvested, no recorded run)
clementpoiret/Perceiver_MNIST/utils/perceiver.py
pointer only (licence: NONE) · get_code("17bf6227f2dcfc36")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, clementpoiret/Perceiver_MNIST when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Perceiver Not yet run
unverified (harvested, no recorded run)
sooheon/perceiver-jax/perceiver_jax/perceiver_jax.py
code served (permissive licence) · get_code("814bc971d852a52e")
How it ran

no invocation recorded.

Perceiver Not yet run
unverified (harvested, no recorded run)
Xianchao-Wu/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("d5ad02791f6342b3")
How it ran

no invocation recorded.

Perceiver Not yet run
unverified (harvested, no recorded run)
deepmind/deepmind-research/perceiver/perceiver.py
code served (permissive licence) · get_code("5f128411733f3d4d")
How it ran

no invocation recorded.

PreNorm Not yet run
unverified (harvested, no recorded run)
clementpoiret/Perceiver_MNIST/utils/perceiver.py
pointer only (licence: NONE) · get_code("5a52aee4bb4c2cf2")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, clementpoiret/Perceiver_MNIST when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

ReZero Not yet run
unverified (harvested, no recorded run)
sooheon/perceiver-jax/perceiver_jax/perceiver_jax.py
code served (permissive licence) · get_code("41eca8c534139e86")
How it ran

no invocation recorded.

cache_fn Not yet run
unverified (harvested, no recorded run)
Xianchao-Wu/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("b88fa1918b8c2d8e")
How it ran

no invocation recorded. this sample carries JSON test cases instead of a setup block; get_reference_implementation serves them with the level they back.

cache_fn Not yet run
unverified (harvested, no recorded run)
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("e56401f8aa027f99")
How it ran

no invocation recorded.

default Not yet run
unverified (harvested, no recorded run)
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("2192232b9e8a1557")
How it ran

no invocation recorded. this sample carries JSON test cases instead of a setup block; get_reference_implementation serves them with the level they back.

default Not yet run
unverified (harvested, no recorded run)
Xianchao-Wu/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("9573db9d1cb8488b")
How it ran

no invocation recorded. this sample carries JSON test cases instead of a setup block; get_reference_implementation serves them with the level they back.

exists Not yet run
unverified (harvested, no recorded run)
lucidrains/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("eb158ba777c1217c")
How it ran

no invocation recorded. this sample carries JSON test cases instead of a setup block; get_reference_implementation serves them with the level they back.

exists Not yet run
unverified (harvested, no recorded run)
Xianchao-Wu/perceiver-pytorch/perceiver_pytorch/perceiver_pytorch.py
code served (permissive licence) · get_code("3f47658c584c3676")
How it ran

no invocation recorded. this sample carries JSON test cases instead of a setup block; get_reference_implementation serves them with the level they back.

fourier_encode Not yet run
unverified (harvested, no recorded run)
Rishit-dagli/Perceiver/perceiver/perceiver.py
code served (permissive licence) · get_code("81535522703627e2")
How it ran

no invocation recorded.

Also present in other papers' code

Of 55 samples lifted from this paper's linked repositories, 55 carry a function identity and 0 do not. Of the 50 distinct functions with an identity, 11 are also present in other papers whose code we hold (5 with at least one paper dated earlier, 2 the same month, 11 later) and 39 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
exists
brick of this paper's assembly: helper
486 other papers
35 dated earlier, 5 the same month, 444 later, 2 undated · 78 only as an assembly brick
1603.08155 (2016-03)
Perceptual Losses for Real-Time Style Transfer and Super-Resolution
1611.07004 (2016-11)
Image-to-Image Translation with Conditional Adversarial Networks
1804.04637 (2018-04)
EMBER: An Open Dataset for Training Static PE Malware Machine Learning Models
and 483 more not named here
default
brick of this paper's assembly: helper
163 other papers
14 dated earlier, 4 the same month, 144 later, 1 undated · 27 only as an assembly brick
1603.08155 (2016-03)
Perceptual Losses for Real-Time Style Transfer and Super-Resolution
1810.00825 (2018-10)
Set Transformer: A Framework for Attention-based Permutation-Invariant Neural Networks
1907.01470 (2019-07)
Augmenting Self-attention with Persistent Memory
and 160 more not named here
GEGLU
brick of this paper's assembly: helper
11 other papers
2 dated earlier, 9 later · 5 only as an assembly brick
1706.03762 (2017-06)
Attention Is All You Need
2102.05095 (2021-02)
Is Space-Time Attention All You Need for Video Understanding?
2107.14795 (2021-07)
Perceiver IO: A General Architecture for Structured Inputs & Outputs
and 8 more not named here
cache_fn
brick of this paper's assembly: helper
9 other papers
1 dated earlier, 8 later · 2 only as an assembly brick
1810.00825 (2018-10)
Set Transformer: A Framework for Attention-based Permutation-Invariant Neural Networks
2107.14795 (2021-07)
Perceiver IO: A General Architecture for Structured Inputs & Outputs
2205.10093 (2022-05)
Visual Concepts Tokenization
and 6 more not named here
PreNorm
brick of this paper's assembly: helper
4 other papers
4 later · 2 only as an assembly brick
2107.14795 (2021-07)
Perceiver IO: A General Architecture for Structured Inputs & Outputs
2210.11170 (2022-10)
Coordinates Are NOT Lonely -Codebook Prior Helps Implicit Neural 3D Representations
2303.15555 (2023-03)
Object Discovery from Motion-Guided Tokens
and 1 more not named here
cache_fn
brick of this paper's assembly: helper
4 other papers
4 later
2311.09115 (2023-11)
HEALNet: Multimodal Fusion for Heterogeneous Biomedical Data
aaai_27766 (2024 (year only))
Inducing Point Operator Transformer: A Flexible and Scalable Architecture for Solving PDEs
2409.12121 (2024-09)
WMCodec: End-to-End Neural Speech Codec with Deep Watermarking for Authenticity Verification
and 1 more not named here
default
brick of this paper's assembly: helper
3 other papers
3 later
2305.13607 (2023-05)
Not All Image Regions Matter: Masked Vector Quantization for Autoregressive Image Generation
2401.01885 (2024-01)
From Audio to Photoreal Embodiment: Synthesizing Humans in Conversations
2410.18712 (2024-10)
Retrieval-Augmented Diffusion Models for Time Series Forecasting
FeedForward
brick of this paper's assembly: helper, model
2 other papers
1 dated earlier, 1 later
2102.05095 (2021-02)
Is Space-Time Attention All You Need for Video Understanding?
2511.11090 (2025-11)
A Space-Time Transformer for Precipitation Nowcasting
fill_diagonal 1 other paper
1 later
2112.06868 (2021-12)
Variational autoencoders in the presence of low-dimensional data: landscape and implicit bias
get_sweep 1 other paper
1 later
2202.07765 (2022-02)
General-purpose, long-context autoregressive modeling with Perceiver AR
l1norm
brick of this paper's assembly: helper
1 other paper
1 later · 1 only as an assembly brick
2212.11972 (2022-12)
Scalable Adaptive Computation for Iterative Generation

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

Biological systems perceive the world by simultaneously processing high-dimensional inputs from modalities as diverse as vision, audition, touch, proprioception, etc. The perception models used in deep learning on the other hand are designed for individual modalities, often relying on domainspecific assumptions such as the local grid structures exploited by virtually all existing vision models. These priors introduce helpful inductive biases, but also lock models to individual modalities. In this paper we introduce the Perceiver -a model that builds upon Transformers and hence makes few architectural assumptions about the relationship between its inputs, but that also scales to hundreds of thousands of inputs, like ConvNets. The model leverages an asymmetric attention mechanism to iteratively distill inputs into a tight latent bottleneck, allowing it to scale to handle very large inputs. We show that this architecture is competitive with or outperforms strong, specialized models on classification tasks across various modalities: images, point clouds, audio, video, and video+audio. The Perceiver obtains performance comparable to ResNet-50 and ViT on ImageNet without 2D convolutions by directly attending to 50,000 pixels. It is also competitive in all modalities in AudioSet.

For agents

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

get_harvested_code_for_paper("2103.03206")
get_code_for_paper("2103.03206")
have("2103.03206")

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

curl https://syntology.ai/api/ran/2103.03206.json

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

[![Syntology run record](https://syntology.ai/api/ran/2103.03206.svg)](https://syntology.ai/paper/2103.03206)

Connect an agent — have() is free.