SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2301.04104 · 2023

Mastering Diverse Domains through World Models

Jimmy Ba, Danijar Hafner, Timothy Lillicrap, Jurgis Pasukonis

arXiv · PDF · Open in the Atlas

Code that ran

21 of 34 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 16 have no contract verdict; 5 failures of our instrument; 13 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/2301.04104.json · badge: /api/ran/2301.04104.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/dreamerv3-torch community
Papers with Code archive
21 ran · 13 no recorded run
FunctionStatusWhere it lives
Bernoulli Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("e681676cbd727836")
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 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
from torch.nn import functional as F
from torch.distributions import Independent, Normal

class Bernoulli:

    def __init__(self, dist=None):
        super().__init__()
        self._dist = dist
        self.mean = dist.mean

    def __getattr__(self, name):
        return getattr(self._dist, name)

    def entropy(self):
        return self._dist.entropy()

    def mode(self):
        _mode = torch.round(self._dist.mean)
        return _mode.detach() + self._dist.mean - self._dist.mean.detach()

    def sample(self, sample_shape=()):
        return self._dist.rsample(sample_shape)

    def log_prob(self, x):
        _logits = self._dist.base_dist.logits
        log_probs0 = -F.softplus(_logits)
        log_probs1 = -F.softplus(-_logits)
        return torch.sum(log_probs0 * (1 - x) + log_probs1 * x, -1)


# Create a mock distribution with logits
base_dist = Normal(loc=torch.zeros(2, 4), scale=torch.ones(2, 4))
base_dist.logits = torch.randn(2, 4)
dist = Independent(base_dist, 1)

INIT_ARGS = {
    'dist': dist
}

ARGS = {
    'x': torch.rand(2, 4)
}

METHOD = 'log_prob'

# Construct and call
instance = Bernoulli(**INIT_ARGS)
result = getattr(instance, METHOD)(**ARGS)
print(result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ContDist Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("176de01fbad3023e")
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 1, dtype family float, shape (3,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.distributions as dist

INIT_ARGS = {
    'dist': dist.Normal(loc=torch.tensor(0.0), scale=torch.tensor(1.0)),
    'absmax': torch.tensor(2.0)
}

ARGS = {
    'sample_shape': (3,)
}

METHOD = 'sample'

if __name__ == '__main__':
    from torch.distributions import Normal
    
    class ContDist:
        def __init__(self, dist=None, absmax=None):
            super().__init__()
            self._dist = dist
            self.mean = dist.mean
            self.absmax = absmax

        def __getattr__(self, name):
            return getattr(self._dist, name)

        def entropy(self):
            return self._dist.entropy()

        def mode(self):
            out = self._dist.mean
            if self.absmax is not None:
                out *= (self.absmax / torch.clip(torch.abs(out), min=self.absmax)).detach()
            return out

        def sample(self, sample_shape=()):
            out = self._dist.rsample(sample_shape)
            if self.absmax is not None:
                out *= (self.absmax / torch.clip(torch.abs(out), min=self.absmax)).detach()
            return out

        def log_prob(self, x):
            return self._dist.log_prob(x)
    
    cont_dist = ContDist(**INIT_ARGS)
    result = cont_dist.sample(**ARGS)
    print(result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 1, dtype family float, shape (3,) — shape (3,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Conv2dSamePad Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("ffedbf10704bc84d")
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 4, dtype family float, shape (2, 4, 5, 7) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as torch_nn
import math

class Conv2dSamePad(torch_nn.Conv2d):

    def calc_same_pad(self, i, k, s, d):
        return max((math.ceil(i / s) - 1) * s + (k - 1) * d + 1 - i, 0)

    def forward(self, x):
        ih, iw = x.size()[-2:]
        pad_h = self.calc_same_pad(i=ih, k=self.kernel_size[0], s=self.stride[0], d=self.dilation[0])
        pad_w = self.calc_same_pad(i=iw, k=self.kernel_size[1], s=self.stride[1], d=self.dilation[1])
        if pad_h > 0 or pad_w > 0:
            x = torch.nn.functional.pad(x, [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2])
        ret = torch.nn.functional.conv2d(x, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups)
        return ret


INIT_ARGS = {
    'in_channels': 3,
    'out_channels': 4,
    'kernel_size': 3,
    'stride': 1,
    'padding': 0,
    'dilation': 1,
    'groups': 1,
    'bias': True,
}

ARGS = {
    'x': torch.randn(2, 3, 5, 7),
}

if __name__ == '__main__':
    conv = Conv2dSamePad(**INIT_ARGS)
    output = conv(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 4, 5, 7) — shape (2, 4, 5, 7) 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)
ConvEncoder Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("8097025bbaf735ac")
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; 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 numpy as np
import torch
from torch import nn
from torch.nn import functional as F
import math
import types as _types

class Conv2dSamePad(torch.nn.Conv2d):

    def calc_same_pad(self, i, k, s, d):
        return max((math.ceil(i / s) - 1) * s + (k - 1) * d + 1 - i, 0)

    def forward(self, x):
        ih, iw = x.size()[-2:]
        pad_h = self.calc_same_pad(i=ih, k=self.kernel_size[0], s=self.stride[0], d=self.dilation[0])
        pad_w = self.calc_same_pad(i=iw, k=self.kernel_size[1], s=self.stride[1], d=self.dilation[1])
        if pad_h > 0 or pad_w > 0:
            x = F.pad(x, [pad_w // 2, pad_w - pad_w // 2, pad_h // 2, pad_h - pad_h // 2])
        ret = F.conv2d(x, self.weight, self.bias, self.stride, self.padding, self.dilation, self.groups)
        return ret

class ImgChLayerNorm(nn.Module):

    def __init__(self, ch, eps=0.001):
        super(ImgChLayerNorm, self).__init__()
        self.norm = torch.nn.LayerNorm(ch, eps=eps)

    def forward(self, x):
        x = x.permute(0, 2, 3, 1)
        x = self.norm(x)
        x = x.permute(0, 3, 1, 2)
        return x

def weight_init(m):
    if isinstance(m, nn.Linear):
        in_num = m.in_features
        out_num = m.out_features
        denoms = (in_num + out_num) / 2.0
        scale = 1.0 / denoms
        std = np.sqrt(scale) / 0.8796256610342398
        nn.init.trunc_normal_(m.weight.data, mean=0.0, std=std, a=-2.0 * std, b=2.0 * std)
        if hasattr(m.bias, 'data'):
            m.bias.data.fill_(0.0)
    elif isinstance(m, nn.Conv2d) or isinstance(m, nn.ConvTranspose2d):
        space = m.kernel_size[0] * m.kernel_size[1]
        in_num = space * m.in_channels
        out_num = space * m.out_channels
        denoms = (in_num + out_num) / 2.0
        scale = 1.0 / denoms
        std = np.sqrt(scale) / 0.8796256610342398
        nn.init.trunc_normal_(m.weight.data, mean=0.0, std=std, a=-2.0 * std, b=2.0 * std)
        if hasattr(m.bias, 'data'):
            m.bias.data.fill_(0.0)
    elif isinstance(m, nn.LayerNorm):
        m.weight.data.fill_(1.0)
        if hasattr(m.bias, 'data'):
            m.bias.data.fill_(0.0)

class ConvEncoder(nn.Module):

    def __init__(self, input_shape, depth=32, act='SiLU', norm=True, kernel_size=4, minres=4):
        super(ConvEncoder, self).__init__()
        act = getattr(torch.nn, act)
        h, w, input_ch = input_shape
        stages = int(np.log2(h) - np.log2(minres))
        in_dim = input_ch
        out_dim = depth
        layers = []
        for i in range(stages):
            layers.append(Conv2dSamePad(in_channels=in_dim, out_channels=out_dim, kernel_size=kernel_size, stride=2, bias=False))
            if norm:
                layers.append(ImgChLayerNorm(out_dim))
            layers.append(act())
            in_dim = out_dim
            out_dim *= 2
            h, w = (h // 2, w // 2)
        self.outdim = out_dim // 2 * h * w
        self.layers = nn.Sequential(*layers)
        self.layers.apply(weight_init)

    def forward(self, obs):
        obs -= 0.5
        x = obs.reshape((-1,) + tuple(obs.shape[-3:]))
        x = x.permute(0, 3, 1, 2)
        x = self.layers(x)
        x = x.reshape([x.shape[0], np.prod(x.shape[1:])])
        return x.reshape(list(obs.shape[:-3]) + [x.shape[-1]])


INIT_ARGS = {
    'input_shape': (16, 12, 3),
    'depth': 8,
    'act': 'SiLU',
    'norm': True,
    'kernel_size': 4,
    'minres': 4,
}

ARGS = {
    'obs': torch.randn(2, 16, 12, 3),
}

if __name__ == '__main__':
    encoder = ConvEncoder(**INIT_ARGS)
    output = encoder(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output dtype: {output.dtype}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
DiscDist Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("40c192080c59445f")
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 1, dtype family float, shape (2,) as a later run recorded it. 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, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch.nn import functional as F

def symexp(x):
    return torch.sign(x) * (torch.exp(torch.abs(x)) - 1.0)

def symlog(x):
    return torch.sign(x) * torch.log(torch.abs(x) + 1.0)

class DiscDist:

    def __init__(self, logits, low=-20.0, high=20.0, transfwd=symlog, transbwd=symexp, device='cpu'):
        self.logits = logits
        self.probs = torch.softmax(logits, -1)
        self.buckets = torch.linspace(low, high, steps=255, device=device)
        self.width = (self.buckets[-1] - self.buckets[0]) / 255
        self.transfwd = transfwd
        self.transbwd = transbwd

    def mean(self):
        _mean = self.probs * self.buckets
        return self.transbwd(torch.sum(_mean, dim=-1, keepdim=True))

    def mode(self):
        _mode = self.probs * self.buckets
        return self.transbwd(torch.sum(_mode, dim=-1, keepdim=True))

    def log_prob(self, x):
        x = self.transfwd(x)
        below = torch.sum((self.buckets <= x[..., None]).to(torch.int32), dim=-1) - 1
        above = len(self.buckets) - torch.sum((self.buckets > x[..., None]).to(torch.int32), dim=-1)
        below = torch.clip(below, 0, len(self.buckets) - 1)
        above = torch.clip(above, 0, len(self.buckets) - 1)
        equal = below == above
        dist_to_below = torch.where(equal, 1, torch.abs(self.buckets[below] - x))
        dist_to_above = torch.where(equal, 1, torch.abs(self.buckets[above] - x))
        total = dist_to_below + dist_to_above
        weight_below = dist_to_above / total
        weight_above = dist_to_below / total
        target = F.one_hot(below, num_classes=len(self.buckets)) * weight_below[..., None] + F.one_hot(above, num_classes=len(self.buckets)) * weight_above[..., None]
        log_pred = self.logits - torch.logsumexp(self.logits, -1, keepdim=True)
        target = target.squeeze(-2)
        return (target * log_pred).sum(-1)

    def log_prob_target(self, target):
        log_pred = self.logits - torch.logsumexp(self.logits, -1, keepdim=True)
        return (target * log_pred).sum(-1)


INIT_ARGS = {
    'logits': torch.randn(2, 255),
    'low': -20.0,
    'high': 20.0,
    'transfwd': symlog,
    'transbwd': symexp,
    'device': 'cpu'
}

ARGS = {
    'x': torch.randn(2, 1)
}

METHOD = 'log_prob'

if __name__ == '__main__':
    dist = DiscDist(**INIT_ARGS)
    result = dist.log_prob(**ARGS)
    print("Result shape:", result.shape)
    print("Result:", result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Every Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("b8806e5208586490")
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 int, ndim 0, dtype family int, shape () as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

class Every:

    def __init__(self, every):
        self._every = every
        self._last = None

    def __call__(self, step):
        if not self._every:
            return 0
        if self._last is None:
            self._last = step
            return 1
        count = int((step - self._last) / self._every)
        self._last += self._every * count
        return count


INIT_ARGS = {
    'every': 5
}

ARGS = {
    'step': 10
}

instance = Every(**INIT_ARGS)
result = instance(**ARGS)
print(f"Result: {result}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type int, ndim 0, dtype family int, shape () — shape () as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
GRUCell Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("b277728e34839765")
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 tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import nn

class GRUCell(nn.Module):

    def __init__(self, inp_size, size, norm=True, act=torch.tanh, update_bias=-1):
        super(GRUCell, self).__init__()
        self._inp_size = inp_size
        self._size = size
        self._act = act
        self._update_bias = update_bias
        self.layers = nn.Sequential()
        self.layers.add_module('GRU_linear', nn.Linear(inp_size + size, 3 * size, bias=False))
        if norm:
            self.layers.add_module('GRU_norm', nn.LayerNorm(3 * size, eps=0.001))

    @property
    def state_size(self):
        return self._size

    def forward(self, inputs, state):
        state = state[0]
        parts = self.layers(torch.cat([inputs, state], -1))
        reset, cand, update = torch.split(parts, [self._size] * 3, -1)
        reset = torch.sigmoid(reset)
        cand = self._act(reset * cand)
        update = torch.sigmoid(update + self._update_bias)
        output = update * cand + (1 - update) * state
        return (output, [output])


INIT_ARGS = {
    'inp_size': 4,
    'size': 3,
    'norm': True,
    'act': torch.tanh,
    'update_bias': -1
}

ARGS = {
    'inputs': torch.randn(2, 4),
    'state': [torch.randn(2, 3)]
}

if __name__ == '__main__':
    cell = GRUCell(**INIT_ARGS)
    output, next_state = cell(**ARGS)
    print("Output shape:", output.shape)
    print("Next state shape:", next_state[0].shape)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ImgChLayerNorm Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("91cc12a004a18d4b")
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 4, dtype family float, shape (2, 4, 3, 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 torch
from torch import nn

class ImgChLayerNorm(nn.Module):

    def __init__(self, ch, eps=0.001):
        super(ImgChLayerNorm, self).__init__()
        self.norm = torch.nn.LayerNorm(ch, eps=eps)

    def forward(self, x):
        x = x.permute(0, 2, 3, 1)
        x = self.norm(x)
        x = x.permute(0, 3, 1, 2)
        return x


INIT_ARGS = {
    'ch': 4,
    'eps': 0.001
}

ARGS = {
    'x': torch.randn(2, 4, 3, 5)
}

if __name__ == '__main__':
    model = ImgChLayerNorm(**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
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 4, 3, 5) — shape (2, 4, 3, 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)
MSEDist Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("8addd6348ab08f08")
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; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# INIT_ARGS for MSEDist.__init__(mode, agg='sum')
# mode should be a tensor representing the predicted/target mode
# agg is the aggregation method

mode_tensor = torch.randn(2, 3, 4, 5)  # batch=2, seq=3, height=4, width=5

INIT_ARGS = {
    'mode': mode_tensor,
    'agg': 'sum'
}

# ARGS for MSEDist.log_prob(value)
# value should have the same shape as mode

value_tensor = torch.randn(2, 3, 4, 5)  # same shape as mode

ARGS = {
    'value': value_tensor
}

METHOD = 'log_prob'
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Once Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("72e56ebf38efd993")
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 bool, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# Once.__init__ takes no parameters (only self)
INIT_ARGS = {}

# Once.__call__ takes no parameters (only self)
ARGS = {}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float — no output shape is stated: observe_signature.py could not observe it (no_inputs)
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)
OneHotDist Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("8fe729e5b488314a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# INIT_ARGS: parameters for OneHotDist.__init__
# The class accepts logits OR probs, and an optional unimix_ratio
# We'll test with logits and a non-zero unimix_ratio to exercise the mixing logic
INIT_ARGS = {
    'logits': torch.randn(2, 5),  # batch_size=2, num_categories=5
    'probs': None,
    'unimix_ratio': 0.01
}

# ARGS: parameters for the sample method (the main computation method)
# sample() takes sample_shape and seed
ARGS = {
    'sample_shape': (3,),  # draw 3 samples
    'seed': None
}

METHOD = 'sample'
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Random Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("60858ba7201fa411")
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 tuple, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import nn
from torch import distributions as torchd
from torch.nn import functional as F
import types as _types

class OneHotDist(torchd.one_hot_categorical.OneHotCategorical):

    def __init__(self, logits=None, probs=None, unimix_ratio=0.0):
        if logits is not None and unimix_ratio > 0.0:
            probs = F.softmax(logits, dim=-1)
            probs = probs * (1.0 - unimix_ratio) + unimix_ratio / probs.shape[-1]
            logits = torch.log(probs)
            super().__init__(logits=logits, probs=None)
        else:
            super().__init__(logits=logits, probs=probs)

    def mode(self):
        _mode = F.one_hot(torch.argmax(super().logits, axis=-1), super().logits.shape[-1])
        return _mode.detach() + super().logits - super().logits.detach()

    def sample(self, sample_shape=(), seed=None):
        if seed is not None:
            raise ValueError('need to check')
        sample = super().sample(sample_shape).detach()
        probs = super().probs
        while len(probs.shape) < len(sample.shape):
            probs = probs[None]
        sample += probs - probs.detach()
        return sample

class Random(nn.Module):

    def __init__(self, config, act_space):
        super(Random, self).__init__()
        self._config = config
        self._act_space = act_space

    def actor(self, feat):
        if self._config.actor['dist'] == 'onehot':
            return tools.OneHotDist(torch.zeros(self._config.num_actions, device=self._config.device).repeat(self._config.envs, 1))
        else:
            return torchd.independent.Independent(torchd.uniform.Uniform(torch.tensor(self._act_space.low, device=self._config.device).repeat(self._config.envs, 1), torch.tensor(self._act_space.high, device=self._config.device).repeat(self._config.envs, 1)), 1)

    def train(self, start, context, data):
        return (None, {})

tools = _types.SimpleNamespace(OneHotDist=OneHotDist)

class MockActSpace:
    def __init__(self):
        self.low = [-1.0, -1.0, -1.0]
        self.high = [1.0, 1.0, 1.0]

config = _types.SimpleNamespace(
    num_actions=5,
    envs=2,
    device='cpu',
    actor={'dist': 'onehot'}
)

act_space = MockActSpace()

INIT_ARGS = {
    'config': config,
    'act_space': act_space
}

ARGS = {
    'start': torch.zeros(2, 4),
    'context': torch.zeros(2, 8),
    'data': torch.zeros(2, 3)
}

METHOD = 'train'

if __name__ == '__main__':
    model = Random(**INIT_ARGS)
    result = model.train(**ARGS)
    print("Result:", result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, dtype family float — a tuple 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
SafeTruncatedNormal Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("81392766dd9e5f90")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# INIT_ARGS for SafeTruncatedNormal.__init__(loc, scale, low, high, clip=1e-06, mult=1)
INIT_ARGS = {
    'loc': torch.tensor([0.0, 1.0, -0.5]),
    'scale': torch.tensor([1.0, 0.5, 2.0]),
    'low': -3.0,
    'high': 3.0,
    'clip': 1e-06,
    'mult': 1
}

# ARGS for SafeTruncatedNormal.sample(sample_shape)
ARGS = {
    'sample_shape': torch.Size([2, 4])
}

METHOD = 'sample'
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
SymlogDist Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("b88fb85f59d2ee19")
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 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

def symexp(x):
    return torch.sign(x) * (torch.exp(torch.abs(x)) - 1.0)

def symlog(x):
    return torch.sign(x) * torch.log(torch.abs(x) + 1.0)

class SymlogDist:

    def __init__(self, mode, dist='mse', agg='sum', tol=1e-08):
        self._mode = mode
        self._dist = dist
        self._agg = agg
        self._tol = tol

    def mode(self):
        return symexp(self._mode)

    def mean(self):
        return symexp(self._mode)

    def log_prob(self, value):
        assert self._mode.shape == value.shape
        if self._dist == 'mse':
            distance = (self._mode - symlog(value)) ** 2.0
            distance = torch.where(distance < self._tol, 0, distance)
        elif self._dist == 'abs':
            distance = torch.abs(self._mode - symlog(value))
            distance = torch.where(distance < self._tol, 0, distance)
        else:
            raise NotImplementedError(self._dist)
        if self._agg == 'mean':
            loss = distance.mean(list(range(len(distance.shape)))[2:])
        elif self._agg == 'sum':
            loss = distance.sum(list(range(len(distance.shape)))[2:])
        else:
            raise NotImplementedError(self._agg)
        return -loss


INIT_ARGS = {
    'mode': torch.randn(2, 4),
    'dist': 'mse',
    'agg': 'sum',
    'tol': 1e-08,
}

ARGS = {
    'value': torch.randn(2, 4),
}

METHOD = 'log_prob'

if __name__ == '__main__':
    dist = SymlogDist(**INIT_ARGS)
    result = dist.log_prob(**ARGS)
    print("Result shape:", result.shape)
    print("Result:", result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
TanhBijector Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("ccb38fe3b67c3f47")
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; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from torch import distributions as torchd

class TanhBijector(torchd.Transform):

    def __init__(self, validate_args=False, name='tanh'):
        super().__init__()

    def _forward(self, x):
        return torch.tanh(x)

    def _inverse(self, y):
        y = torch.where(torch.abs(y) <= 1.0, torch.clamp(y, -0.99999997, 0.99999997), y)
        y = torch.atanh(y)
        return y

    def _forward_log_det_jacobian(self, x):
        log2 = torch.math.log(2.0)
        return 2.0 * (log2 - x - torch.softplus(-2.0 * x))


INIT_ARGS = {
    'validate_args': False,
    'name': 'tanh'
}

METHOD = '_forward'

ARGS = {
    'x': torch.randn(2, 4)
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Until Ran
ran on a synthesized input
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("3d7ec6f83d9cefcf")
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 bool, ndim 0, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# INIT_ARGS for Until.__init__(until)
INIT_ARGS = {
    'until': 10
}

# ARGS for Until.__call__(step)
ARGS = {
    'step': 5
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type bool, ndim 0, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
lambda_return Ran
ran; our contract draft was wrong, not the code
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("3d3a2f5348d90046")
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 tuple, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create small tensors with distinct dimensions
batch_size = 2
time_steps = 3
value_dim = 4

# reward: shape (time_steps, batch_size, value_dim)
reward = torch.randn(time_steps, batch_size, value_dim)

# value: same shape as reward
value = torch.randn(time_steps, batch_size, value_dim)

# pcont: can be scalar (int/float) or tensor matching reward shape
pcont = 0.99

# bootstrap: tensor matching value[-1] shape, or None
bootstrap = None

# lambda_: scalar between 0 and 1
lambda_ = 0.95

# axis: which axis to process (0 in this case since reward/value already in time-first format)
axis = 0

ARGS = {
    "reward": reward,
    "value": value,
    "pcont": pcont,
    "bootstrap": bootstrap,
    "lambda_": lambda_,
    "axis": axis,
}
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 tuple, ndim 3, dtype family float — a tuple 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
symexp Ran
ran; our contract draft was wrong, not the code
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("fd421307a7e4cede")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function symexp takes a single parameter x and applies:
# torch.sign(x) * (torch.exp(torch.abs(x)) - 1.0)
# So x should be a tensor

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 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
symlog Ran
ran; our contract draft was wrong, not the code
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("3e346b9b00ea8169")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function symlog takes a single parameter x and applies:
# torch.sign(x), torch.log(torch.abs(x) + 1.0)
# So x should be a tensor that supports these operations

x = torch.randn(2, 4)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 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/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("37981558f2b7669b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned 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 ARGS dict with all parameters
ARGS = {
    'tensor': tensor,
    'prefix': 'test'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
uniform_weight_init Ran
ran; our contract draft was wrong, not the code
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("8ab3c3275de4031e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np
from torch import nn

torch.manual_seed(0)

# Create a simple model with various layer types that the function handles
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
        self.conv2d = nn.Conv2d(2, 3, kernel_size=(2, 2))
        self.conv_transpose = nn.ConvTranspose2d(3, 2, kernel_size=(3, 3))
        self.layer_norm = nn.LayerNorm(4)
    
    def forward(self, x):
        return x

model = SimpleModel()

# The function returns an initializer function that takes a module
# We need to pass a scale value
given_scale = 1.0

ARGS = {
    "given_scale": given_scale
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ConvDecoder Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("bd06b32e32cc8750")
How it ran

no invocation recorded.

Dreamer Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("0109f7f0801c59db")
How it ran

no invocation recorded.

ImagBehavior Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("51adcb05639dfb9c")
How it ran

no invocation recorded.

MLP Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("02f11645dd0babdb")
How it ran

no invocation recorded.

MultiDecoder Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("034c39a2bc2a4cdd")
How it ran

no invocation recorded.

MultiEncoder Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("c00542f5f0dde998")
How it ran

no invocation recorded.

Optimizer Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("ad4473538a000c98")
How it ran

no invocation recorded.

Plan2Explore Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("e16338beac400c5e")
How it ran

no invocation recorded.

RSSM Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("795c1fe9da508672")
How it ran

no invocation recorded.

RewardEMA Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("d6d8a159b73de69a")
How it ran

no invocation recorded.

SampleDist Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("c8caad78f3b4ea79")
How it ran

no invocation recorded.

WorldModel Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("536e51c6e29f6a5b")
How it ran

no invocation recorded.

weight_init Not yet run
unverified (harvested, no recorded run)
nm512/dreamerv3-torch/dreamer.py
code served (permissive licence) · get_code("9aaedf6b6e636015")
How it ran

no invocation recorded.

Also present in other papers' code

Of 34 samples lifted from this paper's linked repositories, 34 carry a function identity and 0 do not. Of the 34 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 2 later) and 32 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
symexp
brick of this paper's assembly: helper
5 other papers
1 dated earlier, 4 later
2210.13702 (2022-10)
DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality
2310.05167 (2023-10)
Hieros: Hierarchical Imagination on Structured State Space Sequence World Models
2403.09859 (2024-03)
MAMBA: an Effective World Model Approach for Meta-Reinforcement Learning
and 2 more not named here
symlog
brick of this paper's assembly: helper
5 other papers
1 dated earlier, 4 later
2210.13702 (2022-10)
DeXtreme: Transfer of Agile In-hand Manipulation from Simulation to Reality
2310.05167 (2023-10)
Hieros: Hierarchical Imagination on Structured State Space Sequence World Models
2403.09859 (2024-03)
MAMBA: an Effective World Model Approach for Meta-Reinforcement Learning
and 2 more not named here

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

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

Abstract

Developing a general algorithm that learns to solve tasks across a wide range of applications has been a fundamental challenge in artificial intelligence. Although current reinforcement learning algorithms can be readily applied to tasks similar to what they have been developed for, configuring them for new application domains requires significant human expertise and experimentation. We present DreamerV3, a general algorithm that outperforms specialized methods across over 150 diverse tasks, with a single configuration. Dreamer learns a model of the environment and improves its behavior by imagining future scenarios. Robustness techniques based on normalization, balancing, and transformations enable stable learning across domains. Applied out of the box, Dreamer is the first algorithm to collect diamonds in Minecraft from scratch without human data or curricula. This achievement has been posed as a significant challenge in artificial intelligence that requires exploring farsighted strategies from pixels and sparse rewards in an open world. Our work allows solving challenging control problems without extensive experimentation, making reinforcement learning broadly applicable.

For agents

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

get_harvested_code_for_paper("2301.04104")
get_code_for_paper("2301.04104")
have("2301.04104")

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

curl https://syntology.ai/api/ran/2301.04104.json

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

[![Syntology run record](https://syntology.ai/api/ran/2301.04104.svg)](https://syntology.ai/paper/2301.04104)

Connect an agent — have() is free.