SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2603.18202 · 2026

R2-Dreamer: Redundancy-Reduced World Models without Decoders or Augmentation

Tatsuya Harada, Naoki Morihira, Amal Nahar, Kartik Bharadwaj, Yasuhiro Kato, Akinobu Hayashi

arXiv · PDF · Open in the Atlas

Code that ran

20 of 26 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (13 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 0 violated it and 13 have no contract verdict; 5 failures of our instrument; 6 no recorded run), measured 2026-11-22; 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/2603.18202.json · badge: /api/ran/2603.18202.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
NM512/r2dreamer community
Papers with Code archive
20 ran · 6 no recorded run
FunctionStatusWhere it lives
BlockLinear Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("a6e59c299c248bbe")
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, BlockLinear: 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)

ARGS = {
    'in_ch': 8,
    'out_ch': 12,
    'blocks': 4,
    'outscale': 1.0,
}
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 BlockLinear — 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 BlockLinear 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
Conv2dSamePad Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("71a26f571a92c784")
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, Conv2dSamePad: 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)

# Conv2dSamePad inherits from nn.Conv2d, so we need to provide the same arguments
# that nn.Conv2d expects in its __init__:
# in_channels, out_channels, kernel_size, stride, padding, dilation, groups, bias, padding_mode

ARGS = {
    "in_channels": 3,
    "out_channels": 16,
    "kernel_size": (3, 3),
    "stride": (1, 1),
    "padding": (0, 0),
    "dilation": (1, 1),
    "groups": 1,
    "bias": True,
    "padding_mode": "zeros",
}
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 Conv2dSamePad — 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 Conv2dSamePad 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
ConvDecoder Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("7385b4579a238eaa")
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, ConvDecoder: 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

torch.manual_seed(0)

class Config:
    def __init__(self):
        self.act = 'ReLU'
        self.depth = 4
        self.mults = [1, 2]
        self.bspace = 2
        self.kernel_size = 3
        self.units = 8

config = Config()

deter = 16
flat_stoch = 12
shape = (3, 64, 64)

ARGS = {
    'config': config,
    'deter': deter,
    'flat_stoch': flat_stoch,
    'shape': shape,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ConvDecoder — 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 ConvDecoder 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
ConvEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("e577bedaeb6a10cf")
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, ConvEncoder: 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 config object with required attributes
class Config:
    def __init__(self):
        self.act = 'ReLU'
        self.depth = 8
        self.mults = [1, 2]
        self.kernel_size = 3
        self.norm = True

config = Config()

# input_shape is (height, width, channels)
input_shape = (16, 16, 3)

ARGS = {
    'config': config,
    'input_shape': input_shape,
}
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 ConvEncoder — 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 ConvEncoder 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
Deter Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("07e8edee67f55130")
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, Deter: 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 the arguments for Deter.__init__
deter = 16  # deterministic state dimension
stoch = 8   # stochastic state dimension
act_dim = 4  # action dimension
hidden = 12  # hidden dimension
blocks = 2   # number of blocks
dynlayers = 2  # number of dynamic layers
act = 'SiLU'  # activation function

ARGS = {
    'deter': deter,
    'stoch': stoch,
    'act_dim': act_dim,
    'hidden': hidden,
    'blocks': blocks,
    'dynlayers': dynlayers,
    'act': act,
}
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 Deter — 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 Deter 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
LaProp Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("2a50c1dcb5ccc3be")
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, LaProp: 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 small model with parameters to optimize
model = nn.Linear(4, 3, bias=True)

# Extract parameters from the model
params = list(model.parameters())

# Define all arguments for LaProp
ARGS = {
    'params': params,
    'lr': 0.0004,
    'betas': (0.9, 0.999),
    'eps': 1e-15,
    'weight_decay': 0.0,
    'amsgrad': False,
    'centered': False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type LaProp — 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 LaProp 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
LambdaLayer Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("c5d44529e0a1e72e")
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, LambdaLayer: 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)

# LambdaLayer wraps a callable and calls it in forward()
# We need to provide a lambd parameter that is callable

# Create a simple callable (lambda function or regular function)
lambd = lambda x: x * 2 + 1

ARGS = {
    "lambd": lambd
}
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 LambdaLayer — 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 LambdaLayer output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
MLP Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("036445ac0974af3a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, MLP: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import nn
import types as _types

torch.manual_seed(0)

def symlog(x):
    return torch.sign(x) * torch.log1p(torch.abs(x))

def symexp(x):
    return torch.sign(x) * (torch.exp(torch.abs(x)) - 1)

class Config:
    def __init__(self):
        self.act = 'ReLU'
        self.symlog_inputs = False
        self.device = 'cpu'
        self.layers = 2
        self.units = 8
        self.name = 'mlp'

config = Config()
inp_dim = 4

ARGS = {
    'config': config,
    'inp_dim': inp_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 MLP — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MLP output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
MultiEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("d64050aa5bb98bd6")
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, MultiEncoder: 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 re

torch.manual_seed(0)

class Config:
    def __init__(self, **kwargs):
        for k, v in kwargs.items():
            setattr(self, k, v)

# Create a config object with all necessary attributes
config = Config(
    act='ReLU',
    depth=16,
    mults=[1, 2],
    kernel_size=4,
    norm=True,
    cnn_keys='image',
    mlp_keys='vector',
    cnn=Config(
        act='ReLU',
        depth=16,
        mults=[1, 2],
        kernel_size=4,
        norm=True,
    ),
    mlp=Config(
        act='ReLU',
        layers=2,
        units=32,
        symlog_inputs=False,
        device='cpu',
        name='mlp',
    ),
    layers=2,
    units=32,
    symlog_inputs=False,
    device='cpu',
    name='encoder',
)

# Define shapes dictionary with CNN (3D) and MLP (1D, 2D) observations
shapes = {
    'image': (64, 64, 3),  # CNN: (H, W, C)
    'vector': (8,),         # MLP: (features,)
    'state': (4, 5),        # MLP: (batch, features)
    'is_first': (1,),       # excluded
    'reward': (1,),         # excluded
    'log_other': (2,),      # excluded (starts with log_)
}

ARGS = {
    'config': config,
    'shapes': shapes,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MultiEncoder — 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 MultiEncoder 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
OneHotDist Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("eafa68a58c657f72")
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, OneHotDist: 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.functional as F

torch.manual_seed(0)

# Construct logits tensor: shape (batch_size=2, num_classes=4)
# This is the only required parameter for OneHotDist.__init__
logits = torch.randn(2, 4, dtype=torch.float32)

# unimix_ratio is optional with default 0.0
unimix_ratio = 0.0

ARGS = {
    'logits': logits,
    'unimix_ratio': unimix_ratio,
}
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 OneHotDist — 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 OneHotDist 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
Projector Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("adcca2797b687998")
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, Projector: 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 numpy as np
from torch.nn import init as nn_init

torch.manual_seed(0)

def weight_init_(m, fan_type='in'):
    if isinstance(m, nn.RMSNorm):
        with torch.no_grad():
            m.weight.fill_(1.0)
        return
    weight = getattr(m, 'weight', None)
    if weight is None:
        return
    if weight.numel() == 0:
        return
    in_num, out_num = nn_init._calculate_fan_in_and_fan_out(weight)
    with torch.no_grad():
        fan = {'avg': (in_num + out_num) / 2, 'in': in_num, 'out': out_num}[fan_type]
        std = 1.1368 * np.sqrt(1 / fan)
        nn.init.trunc_normal_(weight, mean=0.0, std=std, a=-2.0 * std, b=2.0 * std)
        bias = getattr(m, 'bias', None)
        if bias is not None:
            bias.fill_(0.0)

class Projector(nn.Module):

    def __init__(self, in_ch1, in_ch2):
        super().__init__()
        self.w = nn.Linear(in_ch1, in_ch2, bias=False)
        self.apply(weight_init_)

    def forward(self, x):
        return self.w(x)

ARGS = {
    'in_ch1': 4,
    'in_ch2': 3,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Projector — 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 Projector 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
RMSNorm2D Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("22baf706399a6921")
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, RMSNorm2D: 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)

# RMSNorm2D inherits from nn.RMSNorm and takes ch (number of channels) as parameter
# Looking at the forward method, it expects a 4D tensor (batch, channels, height, width)
# and permutes to channel-last format for RMSNorm, then permutes back

# Create arguments for RMSNorm2D.__init__
ch = 4  # number of channels
eps = 0.001  # epsilon value (has default)
dtype = None  # dtype (has default)

ARGS = {
    "ch": ch,
    "eps": eps,
    "dtype": dtype,
}
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 RMSNorm2D — 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 RMSNorm2D 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
ReturnEMA Ran
ran on a synthesized input · constructed an object; no result was computed
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("5c037c914e1f3cb9")
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, ReturnEMA: 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)

# Construct arguments for ReturnEMA.__init__
device = 'cpu'
alpha = 0.01

ARGS = {
    'device': device,
    'alpha': alpha,
}
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 ReturnEMA — 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 ReturnEMA 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
compute_global_norm Ran
ran, honoured the contract we drafted
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("933d3d65a94aedc8")
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 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 torch

torch.manual_seed(0)

# The function takes a list of tensors
# It flattens them, concatenates them, and computes the L2 norm
# We need to provide a list of tensors (some can be None)

tensor1 = torch.randn(2, 4)
tensor2 = torch.randn(3, 5)
tensor3 = None  # Test that None values are handled
tensor4 = torch.randn(2, 2)

tensors = [tensor1, tensor2, tensor3, tensor4]

ARGS = {
    "tensors": tensors
}
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 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
compute_rms Ran
ran, honoured the contract we drafted
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("713fa0a86dc57cfe")
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 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 torch

torch.manual_seed(0)

# Create a list of tensors with varying shapes and sizes
tensors = [
    torch.randn(2, 4),
    torch.randn(3, 5),
    torch.randn(6),
    None,  # Test that None values are handled
    torch.randn(2, 3, 4),
]

ARGS = {
    "tensors": tensors,
}
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 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
kl Ran
ran; our contract draft was wrong, not the code
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("485e845ae6cf8417")
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 1, dtype family float, shape (2,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors with distinct shapes for the two logits
# batch_size=2, num_classes=4
logits_left = torch.randn(2, 4)
logits_right = torch.randn(2, 4)

ARGS = {
    "logits_left": logits_left,
    "logits_right": logits_right,
}
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 1, dtype family float, shape (2,) — shape (2,) 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
rpad Ran
ran; our contract draft was wrong, not the code
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("2acbc29cb5da0e5e")
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 5, dtype family float, shape (2, 4, 1, 1, 1) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function rpad takes a tensor x and pads it by unsqueezing dimensions
# It needs:
# - x: a tensor
# - pad: an integer specifying how many dimensions to add

x = torch.randn(2, 4)  # Small tensor: batch=2, features=4
pad = 3  # Number of dimensions to unsqueeze

ARGS = {
    "x": x,
    "pad": pad
}
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 5, dtype family float, shape (2, 4, 1, 1, 1) — shape (2, 4, 1, 1, 1) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
symlog Ran
ran; our contract draft was wrong, not the code
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("b6d53c21daa4c865")
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

torch.manual_seed(0)

# symlog takes a single parameter 'x' which should be a tensor
# The function applies sign, log1p, and abs operations on it
# These operations work on any tensor shape

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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
tensorstats Ran
ran; our contract draft was wrong, not the code
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("a2bc31313fa166c4")
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 dict. 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

torch.manual_seed(0)

# Create a small tensor for testing
tensor = torch.randn(2, 4)

# Define the prefix string
prefix = "test"

# Construct ARGS dict with all parameters
ARGS = {
    "tensor": tensor,
    "prefix": prefix
}
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 dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict 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
to_f32 Ran
ran; our contract draft was wrong, not the code
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("3759515ba2a14bc8")
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

torch.manual_seed(0)

# The function to_f32 takes a single parameter 'x' and converts it to float32
# x should be a tensor (since .to() and dtype are tensor methods)

x = torch.randn(2, 4, dtype=torch.float64)

ARGS = {
    "x": x
}
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
Dreamer Not yet run
unverified (harvested, no recorded run)
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("f8994c7ebcd25232")
How it ran

no invocation recorded.

MLPHead Not yet run
unverified (harvested, no recorded run)
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("d9793fedb6dc1770")
How it ran

no invocation recorded.

MultiDecoder Not yet run
unverified (harvested, no recorded run)
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("21fa896fc01c8e43")
How it ran

no invocation recorded.

RSSM Not yet run
unverified (harvested, no recorded run)
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("c96d50956593ba0a")
How it ran

no invocation recorded.

clip_grad_agc_ Not yet run
unverified (harvested, no recorded run)
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("74c5e22ef1636141")
How it ran

no invocation recorded.

weight_init_ Not yet run
unverified (harvested, no recorded run)
NM512/r2dreamer/dreamer.py
code served (permissive licence) · get_code("c5eabf47d7324ad3")
How it ran

no invocation recorded.

Also present in other papers' code

Of 26 samples lifted from this paper's linked repositories, 26 carry a function identity and 0 do not. Of the 26 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, 1 later) and 24 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
LambdaLayer
brick of this paper's assembly: helper
1 other paper
1 later
2605.01973 (2026-05)
Learn-To-Learn on Arbitrary Textual Conditioning: A Hypernetwork-Driven Meta-Gated LLM
symlog
brick of this paper's assembly: helper
1 other paper
1 dated earlier
1912.12482 (2019-12)
SLM Lab: A Comprehensive Benchmark and Modular Software Framework for Reproducible Deep Reinforcement Learning

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

A central challenge in image-based Model-Based Reinforcement Learning (MBRL) is to learn representations that distill essential information from irrelevant visual details. While promising, reconstruction-based methods often waste capacity on large task-irrelevant regions. Decoder-free methods instead learn robust representations by leveraging Data Augmentation (DA), but reliance on such external regularizers limits versatility. We propose R2-Dreamer, a decoder-free MBRL framework with a self-supervised objective that serves as an internal regularizer, preventing representation collapse without resorting to DA. The core of our method is a redundancy-reduction objective inspired by Barlow Twins, which can be easily integrated into existing frameworks. On DeepMind Control Suite and Meta-World, R2-Dreamer is competitive with strong baselines such as DreamerV3 and TD-MPC2 while training 1.59× faster than DreamerV3, and yields substantial gains on DMC-Subtle with tiny task-relevant objects. These results suggest that an effective internal regularizer can enable versatile, high-performance decoder-free MBRL. Code is available at https://github.com/NM512/r2dreamer.

For agents

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

get_harvested_code_for_paper("2603.18202")
get_code_for_paper("2603.18202")
have("2603.18202")

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

curl https://syntology.ai/api/ran/2603.18202.json

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

[![Syntology run record](https://syntology.ai/api/ran/2603.18202.svg)](https://syntology.ai/paper/2603.18202)

Connect an agent — have() is free.