SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2212.11972 · ICML · 2023

Scalable Adaptive Computation for Iterative Generation

Ting Chen, David Fleet, Allan Jabri

arXiv · PDF · Open in the Atlas

Code that ran

21 of 29 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (9 of them constructed an object rather than computing a result; 8 honoured the contract we drafted, 0 violated it and 10 have no contract verdict; 3 failures of our instrument; 8 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2212.11972.json · badge: /api/ran/2212.11972.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
google-research/pix2seq extension (classifier label)
classifier basis: classifier_2025
12 ran · 5 no recorded run
lucidrains/recurrent-interface-network-pytorch community
Papers with Code archive
8 ran · 3 no recorded run
copy not recorded community
not in the paper's declared repositories
1 ran · 0 no recorded run
FunctionStatusWhere it lives
Attend Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("df4d0ae2244e374f")
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, Attend: 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
import torch.nn.functional as F

torch.manual_seed(0)

# The Attend class is instantiated with dropout, flash, and inverted_attention parameters
# We need to create valid arguments for these parameters

# For the forward pass (which is what will be tested), we need:
# q, k, v tensors and optionally a mask

# Create small tensors for attention mechanism
batch_size = 2
num_heads = 4
seq_len_q = 3
seq_len_k = 5
d_model = 8

# q: (batch, heads, seq_len_q, d_model)
q = torch.randn(batch_size, num_heads, seq_len_q, d_model)

# k: (batch, heads, seq_len_k, d_model)
k = torch.randn(batch_size, num_heads, seq_len_k, d_model)

# v: (batch, heads, seq_len_k, d_model)
v = torch.randn(batch_size, num_heads, seq_len_k, d_model)

# mask: (batch, seq_len_k) - optional, will be expanded to (batch, 1, 1, seq_len_k)
mask = torch.ones(batch_size, seq_len_k, dtype=torch.bool)

ARGS = {
    'dropout': 0.1,
    'flash': False,
    'inverted_attention': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, packaging, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Attend — 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 Attend 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/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("305ed1d536ee99b2")
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. 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)

ARGS = {
    'dim': 64,
    'dim_context': 64,
    'heads': 4,
    'dim_head': 16,
    'norm': False,
    'norm_context': False,
    'time_cond_dim': None,
    'flash': False,
    'qk_norm': False,
    'inverted_attention': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, packaging, 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
DropPath Ran
ran on a synthesized input · constructed an object; no result was computed
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("80161ca59f374d2a")
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, DropPath: 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 small test input tensor with shape [batch, height, width, channels]
# Using distinct sizes: batch=2, height=4, width=3, channels=5
test_input = tf.constant(
    np.random.randn(2, 4, 3, 5).astype(np.float32),
    dtype=tf.float32
)

ARGS = {
    'drop_rate': 0.5,
}
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 DropPath — 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 DropPath 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
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("1e9eca95c178f7c9")
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
from einops.layers.torch import Rearrange

torch.manual_seed(0)

class LayerNorm(nn.Module):

    def __init__(self, dim):
        super().__init__()
        self.gamma = nn.Parameter(torch.ones(dim))
        self.register_buffer('beta', torch.zeros(dim))

    def forward(self, x):
        return F.layer_norm(x, x.shape[-1:], self.gamma, self.beta)

def exists(x):
    return x is not None

class FeedForward(nn.Module):

    def __init__(self, dim, mult=4, time_cond_dim=None):
        super().__init__()
        self.norm = LayerNorm(dim)
        self.time_cond = None
        if exists(time_cond_dim):
            self.time_cond = nn.Sequential(nn.SiLU(), nn.Linear(time_cond_dim, dim * 2), Rearrange('b d -> b 1 d'))
            nn.init.zeros_(self.time_cond[-2].weight)
            nn.init.zeros_(self.time_cond[-2].bias)
        inner_dim = int(dim * mult)
        self.net = nn.Sequential(nn.Linear(dim, inner_dim), nn.GELU(), nn.Linear(inner_dim, dim))

    def forward(self, x, time=None):
        x = self.norm(x)
        if exists(self.time_cond):
            assert exists(time)
            scale, shift = self.time_cond(time).chunk(2, dim=-1)
            x = x * (scale + 1) + shift
        return self.net(x)

ARGS = {
    'dim': 4,
    'mult': 4,
    'time_cond_dim': 3
}
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 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
FeedForwardLayer Ran
ran on a synthesized input · constructed an object; no result was computed
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("b9ad5f9347bcd710")
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, FeedForwardLayer: 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

# Create a small test input tensor
# batch_size=2, sequence_length=1, dim_att=4
test_input = tf.constant([[[0.1, 0.2, 0.3, 0.4]], [[0.5, 0.6, 0.7, 0.8]]], dtype=tf.float32)

# Define ARGS dictionary with all parameters for FeedForwardLayer
ARGS = {
    'dim_att': 4,           # attention/output dimension
    'dim_mlp': 8,           # hidden dimension (must be different from dim_att)
    'drop_units': 0.1,      # dropout rate
    'use_ln': True,         # whether to use layer normalization
    'ln_scale_shift': True  # whether layer norm has learnable scale and shift
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type FeedForwardLayer — 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 FeedForwardLayer 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
LayerNorm Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("d070db73849f9743")
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, LayerNorm: 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': 8
}
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 LayerNorm — 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 LayerNorm 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
LearnedSinusoidalPosEmb Ran
ran on a synthesized input
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("e4a1970252d572fd")
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 2, dtype family float, shape (2, 5) 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 math
import torch
from torch import nn
from einops import rearrange

class LearnedSinusoidalPosEmb(nn.Module):

    def __init__(self, dim):
        super().__init__()
        assert dim % 2 == 0
        half_dim = dim // 2
        self.weights = nn.Parameter(torch.randn(half_dim))

    def forward(self, x):
        x = rearrange(x, 'b -> b 1')
        freqs = x * rearrange(self.weights, 'd -> 1 d') * 2 * math.pi
        fouriered = torch.cat((freqs.sin(), freqs.cos()), dim=-1)
        fouriered = torch.cat((x, fouriered), dim=-1)
        return fouriered


INIT_ARGS = {
    'dim': 4
}

ARGS = {
    'x': torch.randn(2)
}

if __name__ == '__main__':
    model = LearnedSinusoidalPosEmb(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Input shape: {ARGS['x'].shape}")
    print(f"Output shape: {output.shape}")
    print(f"Output:\n{output}")
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 2, dtype family float, shape (2, 5) — shape (2, 5) 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)
LinearAttention Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("f8753f538bdf8090")
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, LinearAttention: 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
from einops import rearrange, reduce, repeat
from einops.layers.torch import Rearrange

torch.manual_seed(0)

class LayerNorm(nn.Module):

    def __init__(self, dim):
        super().__init__()
        self.gamma = nn.Parameter(torch.ones(dim))
        self.register_buffer('beta', torch.zeros(dim))

    def forward(self, x):
        return F.layer_norm(x, x.shape[-1:], self.gamma, self.beta)

class MultiHeadedRMSNorm(nn.Module):

    def __init__(self, dim, heads=1):
        super().__init__()
        self.scale = dim ** 0.5
        self.gamma = nn.Parameter(torch.ones(heads, 1, dim))

    def forward(self, x):
        return F.normalize(x, dim=-1) * self.scale * self.gamma

def exists(x):
    return x is not None

class LinearAttention(nn.Module):

    def __init__(self, dim, heads=4, dim_head=32, norm=False, qk_norm=False, time_cond_dim=None):
        super().__init__()
        hidden_dim = dim_head * heads
        self.scale = dim_head ** (-0.5)
        self.heads = heads
        self.time_cond = None
        if exists(time_cond_dim):
            self.time_cond = nn.Sequential(nn.SiLU(), nn.Linear(time_cond_dim, dim * 2), Rearrange('b d -> b 1 d'))
            nn.init.zeros_(self.time_cond[-2].weight)
            nn.init.zeros_(self.time_cond[-2].bias)
        self.norm = LayerNorm(dim) if norm else nn.Identity()
        self.to_qkv = nn.Linear(dim, hidden_dim * 3, bias=False)
        self.qk_norm = qk_norm
        if qk_norm:
            self.q_norm = MultiHeadedRMSNorm(dim_head, heads)
            self.k_norm = MultiHeadedRMSNorm(dim_head, heads)
        self.to_out = nn.Sequential(nn.Linear(hidden_dim, dim, bias=False), LayerNorm(dim))

    def forward(self, x, time=None):
        h = self.heads
        x = self.norm(x)
        if exists(self.time_cond):
            assert exists(time)
            scale, shift = self.time_cond(time).chunk(2, dim=-1)
            x = x * (scale + 1) + shift
        qkv = self.to_qkv(x).chunk(3, dim=-1)
        q, k, v = map(lambda t: rearrange(t, 'b n (h d) -> b h n d', h=h), qkv)
        if self.qk_norm:
            q = self.q_norm(q)
            k = self.k_norm(k)
        q = q.softmax(dim=-1)
        k = k.softmax(dim=-2)
        q = q * self.scale
        context = torch.einsum('b h n d, b h n e -> b h d e', k, v)
        out = torch.einsum('b h d e, b h n d -> b h n e', context, q)
        out = rearrange(out, 'b h n d -> b n (h d)')
        return self.to_out(out)

# Create a minimal test input
batch_size = 2
seq_len = 3
dim = 8
heads = 2
dim_head = 4
time_cond_dim = 6

x = torch.randn(batch_size, seq_len, dim)
time = torch.randn(batch_size, time_cond_dim)

ARGS = {
    'dim': dim,
    'heads': heads,
    'dim_head': dim_head,
    'norm': False,
    'qk_norm': False,
    'time_cond_dim': time_cond_dim
}
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 LinearAttention — 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 LinearAttention 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
MultiHeadedRMSNorm Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("8535fbe5734cf1d3")
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, MultiHeadedRMSNorm: 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)

# The MultiHeadedRMSNorm.__init__ takes dim and heads parameters
# The forward method takes x and performs:
# - F.normalize(x, dim=-1) -> normalizes along last dimension
# - multiplies by self.scale (a float computed from dim)
# - multiplies by self.gamma (shape: [heads, 1, dim])
# So x should have shape compatible with broadcasting with gamma
# x shape should be [..., dim] where ... can include heads dimension

dim = 4
heads = 3

# Create a test input tensor
# Shape: [batch=2, heads=3, dim=4] to match the gamma parameter shape
x = torch.randn(2, heads, dim)

ARGS = {
    "dim": dim,
    "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 MultiHeadedRMSNorm — 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 MultiHeadedRMSNorm 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
PEG Ran
ran on a synthesized input · constructed an object; no result was computed
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("2090cfddb3b4ea08")
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, PEG: 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
import math
from einops import rearrange

torch.manual_seed(0)

class PEG(nn.Module):

    def __init__(self, dim):
        super().__init__()
        self.ds_conv = nn.Conv2d(dim, dim, 3, padding=1, groups=dim)

    def forward(self, x):
        b, n, d = x.shape
        hw = int(math.sqrt(n))
        x = rearrange(x, 'b (h w) d -> b d h w', h=hw)
        x = self.ds_conv(x)
        x = rearrange(x, 'b d h w -> b (h w) d')
        return x

# Create a PEG instance with dim=4
peg_model = PEG(dim=4)

# Create input tensor: shape (batch=2, n=16, d=4)
# n must be a perfect square for the forward pass to work
# 16 = 4x4, so hw = 4
x = torch.randn(2, 16, 4)

ARGS = {
    'dim': 4
}
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 PEG — 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 PEG 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
add_vis_pos_emb Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("fbc958c1295b5f01")
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 SymbolicTensor. 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

class MockModel:
    def __init__(self):
        self.name = "test_model"
        self.weights = []
    
    def add_weight(self, shape, initializer, name):
        weight = tf.Variable(
            initializer(shape=shape, dtype=tf.float32),
            name=name,
            trainable=True
        )
        self.weights.append(weight)
        return weight

mock_self = MockModel()

ARGS = {
    'self': mock_self,
    'pos_encoding': 'sin_cos',
    'n_rows': 4,
    'n_cols': 3,
    'dim': 8,
    'name_prefix': 'test_prefix',
    'initializer': None,
    'return_only': False,
    'normalization_max': 6.2831852
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SymbolicTensor — a SymbolicTensor 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
get_1d_position_codes Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("cbbd048907025ae1")
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 EagerTensor, ndim 3, dtype family float, shape (1, 8, 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 tensorflow as tf

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

# Define ARGS dict with parameters for get_1d_position_codes
ARGS = {
    "seqlen": 8,
    "out_dim": 4,
    "normalization_max": 6.2831852
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
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, shape (1, 8, 4) — shape (1, 8, 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
get_2d_position_codes Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("554d30594bb18ba4")
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 EagerTensor, ndim 4, dtype family float, shape (1, 4, 3, 8) 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 tensorflow as tf

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

# Define arguments for get_2d_position_codes
ARGS = {
    "height": 4,
    "width": 3,
    "out_dim": 8,
    "normalization_max": 6.2831852
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (1, 4, 3, 8) — shape (1, 4, 3, 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
get_angles Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("c1a5330036a08dfa")
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 EagerTensor, ndim 0, dtype family float, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import tensorflow as tf

# Analyze function usage:
# - pos: cast to tf.float32, multiplied by angle_rates (numeric)
# - i: integer division by 2, used in exponent
# - dim: used as divisor for scaling

# Create small test values
pos = 0  # scalar position
i = 1    # scalar index
dim = 4  # dimension size

ARGS = {
    "pos": pos,
    "i": i,
    "dim": dim
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 0, dtype family float, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_ar_mask Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("508ee551fb33fedc")
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 EagerTensor, ndim 4, dtype family float, shape (1, 1, 8, 8) 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 tensorflow as tf

ARGS = {
    "seq_len": 8,
    "dtype": tf.float32,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (1, 1, 8, 8) — shape (1, 1, 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
get_chunk_ar_mask Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("3e4ac71a6f12be9d")
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 EagerTensor, ndim 4, dtype family float, shape (1, 1, 8, 8) 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 tensorflow as tf

ARGS = {
    "seq_len": 8,
    "chunk_size": 4,
    "dtype": tf.float32
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 4, dtype family float, shape (1, 1, 8, 8) — shape (1, 1, 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
get_layer_config Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("90a86ebf31046958")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 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 re

def get_layer_config(layers_str):
    layer_configs = []
    if layers_str.endswith(')'):
        layers_str += '0'
    if not re.match('^(\\([0-9]+\\)[0-9]+)+$', layers_str):
        raise ValueError(f'Unrecognized layer specification {layers_str}.')
    blocks = re.findall('\\([0-9]+\\)[0-9]+', layers_str)
    for i, block in enumerate(blocks):
        x_layers, l_layers = map(int, re.findall('([0-9]+)', block))
        if i < len(blocks) - 1:
            assert l_layers > 0
        layer_configs.append((x_layers, l_layers))
    return layer_configs

ARGS = {
    'layers_str': '(2)3(4)5(6)0'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
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
get_variable_initializer Ran
ran; our contract draft was wrong, not the code
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("37888d00e9d8c4ab")
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 TruncatedNormal. 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

ARGS = {
    "name": None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type TruncatedNormal — a TruncatedNormal output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a TruncatedNormal 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
kronecker_product Ran
ran, honoured the contract we drafted
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("e37890b1d73b280c")
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 EagerTensor, ndim 2, dtype family float, shape (6, 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 tensorflow as tf

# Create two small matrices for testing the Kronecker product
# Using distinct dimensions to make shapes unambiguous
mat1 = tf.constant([[1.0, 2.0, 3.0],
                    [4.0, 5.0, 6.0]], dtype=tf.float32)  # 2x3 matrix

mat2 = tf.constant([[7.0, 8.0],
                    [9.0, 10.0],
                    [11.0, 12.0]], dtype=tf.float32)  # 3x2 matrix

ARGS = {
    "mat1": mat1,
    "mat2": mat2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EagerTensor, ndim 2, dtype family float, shape (6, 6) — shape (6, 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
positional_encoding Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("7a5b750c9c8d5e81")
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 SymbolicTensor. 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

# Create small tensor inputs for positional_encoding
batch_size = 2
sequence_length = 3
dim = 4

# coords should be shape (bsz, size) - batch of 2, sequence length of 3
coords = tf.constant([[0, 1, 2], [0, 1, 2]], dtype=tf.int32)

ARGS = {
    "coords": coords,
    "dim": dim
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
tensorflow — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SymbolicTensor — a SymbolicTensor 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
suffix_id Ran
ran; our contract draft was wrong, not the code
this paper's copy was not recorded; identical code first harvested from google-research/pix2seq
pointer only · get_code("d1c561539ef11926")
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.

FIT Not yet run
unverified (harvested, no recorded run)
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("e147f983e472692f")
How it ran

no invocation recorded.

MLP Not yet run
unverified (harvested, no recorded run)
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("e2256176baee1426")
How it ran

no invocation recorded.

RIN Not yet run
unverified (harvested, no recorded run)
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("5fdaf6caf1863ed1")
How it ran

no invocation recorded.

RINBlock Not yet run
unverified (harvested, no recorded run)
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("89c11df753e06cb9")
How it ran

no invocation recorded.

Sequential Not yet run
unverified (harvested, no recorded run)
lucidrains/recurrent-interface-network-pytorch/rin_pytorch/rin_pytorch.py
code served (permissive licence) · get_code("d6fde31f78c4502f")
How it ran

no invocation recorded.

TransformerDecoderLayer Not yet run
unverified (harvested, no recorded run)
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("f14aac70f9859dd8")
How it ran

no invocation recorded.

TransformerEncoder Not yet run
unverified (harvested, no recorded run)
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("b019a0895599116d")
How it ran

no invocation recorded.

TransformerEncoderLayer Not yet run
unverified (harvested, no recorded run)
google-research/pix2seq/architectures/transformers.py
code served (permissive licence) · get_code("e730a175de3303f8")
How it ran

no invocation recorded.

Also present in other papers' code

Of 29 samples lifted from this paper's linked repositories, 29 carry a function identity and 0 do not. Of the 29 distinct functions with an identity, 2 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 2 later) and 27 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
LayerNorm
brick of this paper's assembly: model
4 other papers
2 dated earlier, 2 later
2203.07852 (2022-03)
Block-Recurrent Transformers
2204.02311 (2022-04)
PaLM: Scaling Language Modeling with Pathways
2308.00951 (2023-08)
From Sparse to Soft Mixtures of Experts
and 1 more not named here
suffix_id
brick of this paper's assembly: helper
1 other paper
1 later
2305.12689 (2023-05)
FIT: Far-reaching Interleaved Transformers

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

Natural data is redundant yet predominant architectures tile computation uniformly across their input and output space. We propose the Recurrent Interface Networks (RINs), an attention-based architecture that decouples its core computation from the dimensionality of the data, enabling adaptive computation for more scalable generation of high-dimensional data. RINs focus the bulk of computation (i.e. global self-attention) on a set of latent tokens, using cross-attention to read and write (i.e. route) information between latent and data tokens. Stacking RIN blocks allows bottom-up (data to latent) and top-down (latent to data) feedback, leading to deeper and more expressive routing. While this routing introduces challenges, this is less problematic in recurrent computation settings where the task (and routing problem) changes gradually, such as iterative generation with diffusion models. We show how to leverage recurrence by conditioning the latent tokens at each forward pass of the reverse diffusion process with those from prior computation, i.e. latent self-conditioning. RINs yield state-of-the-art pixel diffusion models for image and video generation, scaling to 1024×1024 images without cascades or guidance, while being domain-agnostic and up to 10× more efficient than 2D and 3D U-Nets.

For agents

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

get_harvested_code_for_paper("2212.11972")
get_code_for_paper("2212.11972")
have("2212.11972")

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

curl https://syntology.ai/api/ran/2212.11972.json

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

[![Syntology run record](https://syntology.ai/api/ran/2212.11972.svg)](https://syntology.ai/paper/2212.11972)

Connect an agent — have() is free.