SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.04680 · 2026

Audio ControlNet for Fine-Grained Audio Generation and Editing

Xie Chen, Bowen Zhang, Mingqi Yang, Ziyang Ma, Jianwei Yu, Haina Zhu, Yao Xiao, Xiquan Li

arXiv · PDF · Open in the Atlas

Code that ran

18 of 23 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; 1 honoured the contract we drafted, 0 violated it and 13 have no contract verdict; 4 failures of our instrument; 5 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/2602.04680.json · badge: /api/ran/2602.04680.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
haidog-yaqub/EzAudio community
Papers with Code archive
18 ran · 5 no recorded run
FunctionStatusWhere it lives
AdaLN Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("1999017e042f05f2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 6, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

class AdaLN(nn.Module):

    def __init__(self, dim, ada_mode='ada', r=None, alpha=None):
        super().__init__()
        self.ada_mode = ada_mode
        self.scale_shift_table = None
        if ada_mode == 'ada':
            self.time_ada = nn.Linear(dim, 6 * dim, bias=True)
        elif ada_mode == 'ada_single':
            self.scale_shift_table = nn.Parameter(torch.zeros(6, dim))
        elif ada_mode in ['ada_sola', 'ada_sola_bias']:
            self.lora_a = nn.Linear(dim, r * 6, bias=False)
            self.lora_b = nn.Linear(r * 6, dim * 6, bias=False)
            self.scaling = alpha / r
            if ada_mode == 'ada_sola_bias':
                self.scale_shift_table = nn.Parameter(torch.zeros(6, dim))
        else:
            raise NotImplementedError

    def forward(self, time_token=None, time_ada=None):
        if self.ada_mode == 'ada':
            assert time_ada is None
            B = time_token.shape[0]
            time_ada = self.time_ada(time_token).reshape(B, 6, -1)
        elif self.ada_mode == 'ada_single':
            B = time_ada.shape[0]
            time_ada = time_ada.reshape(B, 6, -1)
            time_ada = self.scale_shift_table[None] + time_ada
        elif self.ada_mode in ['ada_sola', 'ada_sola_bias']:
            B = time_ada.shape[0]
            time_ada_lora = self.lora_b(self.lora_a(time_token)) * self.scaling
            time_ada = time_ada + time_ada_lora
            time_ada = time_ada.reshape(B, 6, -1)
            if self.scale_shift_table is not None:
                time_ada = self.scale_shift_table[None] + time_ada
        else:
            raise NotImplementedError
        return time_ada


INIT_ARGS = {
    'dim': 4,
    'ada_mode': 'ada',
    'r': None,
    'alpha': None,
}

ARGS = {
    'time_token': torch.randn(2, 4),
    'time_ada': None,
}

if __name__ == '__main__':
    model = AdaLN(**INIT_ARGS)
    output = model(**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 3, dtype family float, shape (2, 6, 4) — shape (2, 6, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ApproximateGELU Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("855ca3a1a7d22ee5")
How it ran

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

class ApproximateGELU(nn.Module):

    def __init__(self, dim_in: int, dim_out: int, bias: bool=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out, bias=bias)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.proj(x)
        return x * torch.sigmoid(1.702 * x)


INIT_ARGS = {
    'dim_in': 4,
    'dim_out': 3,
    'bias': True,
}

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

if __name__ == '__main__':
    model = ApproximateGELU(**INIT_ARGS)
    output = model(**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 2, dtype family float, shape (2, 3) — shape (2, 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)
FeedForward Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("e85a8130cccd0e41")
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 torch.nn as nn
import torch.nn.functional as F

class ApproximateGELU(nn.Module):

    def __init__(self, dim_in: int, dim_out: int, bias: bool=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out, bias=bias)

    def forward(self, x: torch.Tensor) -> torch.Tensor:
        x = self.proj(x)
        return x * torch.sigmoid(1.702 * x)

class GEGLU(nn.Module):

    def __init__(self, dim_in: int, dim_out: int, bias: bool=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out * 2, bias=bias)

    def gelu(self, gate: torch.Tensor) -> torch.Tensor:
        if gate.device.type != 'mps':
            return F.gelu(gate)
        return F.gelu(gate.to(dtype=torch.float32)).to(dtype=gate.dtype)

    def forward(self, hidden_states):
        hidden_states = self.proj(hidden_states)
        hidden_states, gate = hidden_states.chunk(2, dim=-1)
        return hidden_states * self.gelu(gate)

class GELU(nn.Module):

    def __init__(self, dim_in: int, dim_out: int, approximate: str='none', bias: bool=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out, bias=bias)
        self.approximate = approximate

    def gelu(self, gate: torch.Tensor) -> torch.Tensor:
        if gate.device.type != 'mps':
            return F.gelu(gate, approximate=self.approximate)
        return F.gelu(gate.to(dtype=torch.float32), approximate=self.approximate).to(dtype=gate.dtype)

    def forward(self, hidden_states):
        hidden_states = self.proj(hidden_states)
        hidden_states = self.gelu(hidden_states)
        return hidden_states

class GESnake(nn.Module):

    def __init__(self, dim_in, dim_out, bias, alpha_trainable=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out * 2, bias=bias)
        self.alpha = nn.Parameter(torch.ones(1, 1, dim_out))
        self.beta = nn.Parameter(torch.ones(1, 1, dim_out))
        self.alpha.requires_grad = alpha_trainable
        self.beta.requires_grad = alpha_trainable

    def forward(self, x):
        x = self.proj(x)
        x, gate = x.chunk(2, dim=-1)
        return x * snake_beta(gate, self.alpha, self.beta)

class Snake(nn.Module):

    def __init__(self, dim_in, dim_out, bias, alpha_trainable=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out, bias=bias)
        self.alpha = nn.Parameter(torch.ones(1, 1, dim_out))
        self.beta = nn.Parameter(torch.ones(1, 1, dim_out))
        self.alpha.requires_grad = alpha_trainable
        self.beta.requires_grad = alpha_trainable

    def forward(self, x):
        x = self.proj(x)
        x = snake_beta(x, self.alpha, self.beta)
        return x

def snake_beta(x, alpha, beta):
    return x + beta * torch.sin(x * alpha).pow(2)

class FeedForward(nn.Module):

    def __init__(self, dim, dim_out=None, mult=4, dropout=0.0, activation_fn='geglu', final_dropout=False, inner_dim=None, bias=True):
        super().__init__()
        if inner_dim is None:
            inner_dim = int(dim * mult)
        dim_out = dim_out if dim_out is not None else dim
        if activation_fn == 'gelu':
            act_fn = GELU(dim, inner_dim, bias=bias)
        elif activation_fn == 'gelu-approximate':
            act_fn = GELU(dim, inner_dim, approximate='tanh', bias=bias)
        elif activation_fn == 'geglu':
            act_fn = GEGLU(dim, inner_dim, bias=bias)
        elif activation_fn == 'geglu-approximate':
            act_fn = ApproximateGELU(dim, inner_dim, bias=bias)
        elif activation_fn == 'snake':
            act_fn = Snake(dim, inner_dim, bias=bias)
        elif activation_fn == 'gesnake':
            act_fn = GESnake(dim, inner_dim, bias=bias)
        else:
            raise NotImplementedError
        self.net = nn.ModuleList([])
        self.net.append(act_fn)
        self.net.append(nn.Dropout(dropout))
        self.net.append(nn.Linear(inner_dim, dim_out, bias=bias))
        if final_dropout:
            self.net.append(nn.Dropout(dropout))

    def forward(self, hidden_states: torch.Tensor) -> torch.Tensor:
        for module in self.net:
            hidden_states = module(hidden_states)
        return hidden_states

INIT_ARGS = {
    'dim': 4,
    'dim_out': 3,
    'mult': 4,
    'dropout': 0.1,
    'activation_fn': 'geglu',
    'final_dropout': False,
    'inner_dim': None,
    'bias': True,
}

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

if __name__ == '__main__':
    model = FeedForward(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 3)")
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)
GEGLU Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("26b300e620311af5")
How it ran

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

class GEGLU(nn.Module):

    def __init__(self, dim_in: int, dim_out: int, bias: bool=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out * 2, bias=bias)

    def gelu(self, gate: torch.Tensor) -> torch.Tensor:
        if gate.device.type != 'mps':
            return F.gelu(gate)
        return F.gelu(gate.to(dtype=torch.float32)).to(dtype=gate.dtype)

    def forward(self, hidden_states):
        hidden_states = self.proj(hidden_states)
        hidden_states, gate = hidden_states.chunk(2, dim=-1)
        return hidden_states * self.gelu(gate)


INIT_ARGS = {
    'dim_in': 4,
    'dim_out': 3,
    'bias': True,
}

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

if __name__ == '__main__':
    model = GEGLU(**INIT_ARGS)
    output = model(**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 2, dtype family float, shape (2, 3) — shape (2, 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)
GELU Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("a4664f5e974677e5")
How it ran

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

class GELU(nn.Module):

    def __init__(self, dim_in: int, dim_out: int, approximate: str='none', bias: bool=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out, bias=bias)
        self.approximate = approximate

    def gelu(self, gate: torch.Tensor) -> torch.Tensor:
        if gate.device.type != 'mps':
            return F.gelu(gate, approximate=self.approximate)
        return F.gelu(gate.to(dtype=torch.float32), approximate=self.approximate).to(dtype=gate.dtype)

    def forward(self, hidden_states):
        hidden_states = self.proj(hidden_states)
        hidden_states = self.gelu(hidden_states)
        return hidden_states


INIT_ARGS = {
    'dim_in': 4,
    'dim_out': 3,
    'approximate': 'none',
    'bias': True
}

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

if __name__ == '__main__':
    model = GELU(**INIT_ARGS)
    output = model(**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 2, dtype family float, shape (2, 3) — shape (2, 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)
GESnake Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("f88c5e27a48552c9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (1, 2, 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.nn as nn

def snake_beta(x, alpha, beta):
    return x + beta * torch.sin(x * alpha).pow(2)

class GESnake(nn.Module):

    def __init__(self, dim_in, dim_out, bias, alpha_trainable=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out * 2, bias=bias)
        self.alpha = nn.Parameter(torch.ones(1, 1, dim_out))
        self.beta = nn.Parameter(torch.ones(1, 1, dim_out))
        self.alpha.requires_grad = alpha_trainable
        self.beta.requires_grad = alpha_trainable

    def forward(self, x):
        x = self.proj(x)
        x, gate = x.chunk(2, dim=-1)
        return x * snake_beta(gate, self.alpha, self.beta)


INIT_ARGS = {
    'dim_in': 4,
    'dim_out': 3,
    'bias': True,
    'alpha_trainable': True,
}

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

if __name__ == '__main__':
    model = GESnake(**INIT_ARGS)
    output = model(**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 3, dtype family float, shape (1, 2, 3) — shape (1, 2, 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)
PE_wrapper Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("f9bb66c0ba81da7e")
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 torch.nn as nn
import torch.nn.functional as F
import math
import warnings

class PositionalConvEmbedding(nn.Module):
    """
    Relative positional embedding used in HuBERT
    """

    def __init__(self, dim=768, kernel_size=128, groups=16):
        super().__init__()
        self.conv = nn.Conv1d(dim, dim, kernel_size=kernel_size, padding=kernel_size // 2, groups=groups, bias=True)
        self.conv = nn.utils.parametrizations.weight_norm(self.conv, name='weight', dim=2)

    def forward(self, x):
        x = self.conv(x)
        x = F.gelu(x[:, :, :-1])
        return x

class SinusoidalPositionalEncoding(nn.Module):

    def __init__(self, dim, length):
        super(SinusoidalPositionalEncoding, self).__init__()
        self.length = length
        self.dim = dim
        self.register_buffer('pe', self._generate_positional_encoding(length, dim))

    def _generate_positional_encoding(self, length, dim):
        pe = torch.zeros(length, dim)
        position = torch.arange(0, length, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, dim, 2).float() * (-math.log(10000.0) / dim))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0)
        return pe

    def forward(self, x):
        x = x + self.pe[:, :x.size(1)]
        return x

def _no_grad_trunc_normal_(tensor, mean, std, a, b):

    def norm_cdf(x):
        return (1.0 + math.erf(x / math.sqrt(2.0))) / 2.0
    if mean < a - 2 * std or mean > b + 2 * std:
        warnings.warn('mean is more than 2 std from [a, b] in nn.init.trunc_normal_. The distribution of values may be incorrect.', stacklevel=2)
    with torch.no_grad():
        l = norm_cdf((a - mean) / std)
        u = norm_cdf((b - mean) / std)
        tensor.uniform_(2 * l - 1, 2 * u - 1)
        tensor.erfinv_()
        tensor.mul_(std * math.sqrt(2.0))
        tensor.add_(mean)
        tensor.clamp_(min=a, max=b)
        return tensor

def trunc_normal_(tensor, mean=0.0, std=1.0, a=-2.0, b=2.0):
    """Fills the input Tensor with values drawn from a truncated
    normal distribution. The values are effectively drawn from the
    normal distribution :math:`\\mathcal{N}(\\text{mean}, \\text{std}^2)`
    with values outside :math:`[a, b]` redrawn until they are within
    the bounds. The method used for generating the random values works
    best when :math:`a \\leq \\text{mean} \\leq b`.
    Args:
        tensor: an n-dimensional `torch.Tensor`
        mean: the mean of the normal distribution
        std: the standard deviation of the normal distribution
        a: the minimum cutoff value
        b: the maximum cutoff value
    Examples:
        >>> w = torch.empty(3, 5)
        >>> nn.init.trunc_normal_(w)
    """
    return _no_grad_trunc_normal_(tensor, mean, std, a, b)

class PE_wrapper(nn.Module):

    def __init__(self, dim=768, method='abs', length=None, **kwargs):
        super().__init__()
        self.method = method
        if method == 'abs':
            self.length = length
            self.abs_pe = nn.Parameter(torch.zeros(1, length, dim))
            trunc_normal_(self.abs_pe, std=0.02)
        elif method == 'conv':
            self.conv_pe = PositionalConvEmbedding(dim=dim, **kwargs)
        elif method == 'sinu':
            self.sinu_pe = SinusoidalPositionalEncoding(dim=dim, length=length)
        elif method == 'none':
            self.id = nn.Identity()
        else:
            raise NotImplementedError

    def forward(self, x):
        if self.method == 'abs':
            _, L, _ = x.shape
            assert L <= self.length
            x = x + self.abs_pe[:, :L, :]
        elif self.method == 'conv':
            x = x + self.conv_pe(x)
        elif self.method == 'sinu':
            x = self.sinu_pe(x)
        elif self.method == 'none':
            x = self.id(x)
        else:
            raise NotImplementedError
        return x


INIT_ARGS = {
    'dim': 4,
    'method': 'abs',
    'length': 8,
}

ARGS = {
    'x': torch.randn(2, 6, 4),
}

if __name__ == '__main__':
    model = PE_wrapper(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shape:", output.shape)
    print("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 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)
PatchEmbed Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("9f2cdbcb2103e1b1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 4, 64) 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 nn

class PatchEmbed(nn.Module):
    """
     Image to Patch Embedding
    """

    def __init__(self, patch_size, in_chans=3, embed_dim=768, input_type='2d'):
        super().__init__()
        self.patch_size = patch_size
        self.input_type = input_type
        if input_type == '2d':
            self.proj = nn.Conv2d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size, bias=True)
        elif input_type == '1d':
            self.proj = nn.Conv1d(in_chans, embed_dim, kernel_size=patch_size, stride=patch_size, bias=True)

    def forward(self, x):
        if self.input_type == '2d':
            B, C, H, W = x.shape
            assert H % self.patch_size == 0 and W % self.patch_size == 0
        elif self.input_type == '1d':
            B, C, H = x.shape
            assert H % self.patch_size == 0
        x = self.proj(x).flatten(2).transpose(1, 2)
        return x


INIT_ARGS = {
    'patch_size': 4,
    'in_chans': 3,
    'embed_dim': 64,
    'input_type': '2d'
}

ARGS = {
    'x': torch.randn(2, 3, 8, 8)
}

if __name__ == '__main__':
    model = PatchEmbed(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Expected shape: (2, 4, 64)")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 4, 64) — shape (2, 4, 64) 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)
PositionalConvEmbedding Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("dd416fb8e3e347c7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 4, 16) 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 nn
import torch.nn.functional as F

class PositionalConvEmbedding(nn.Module):
    """
    Relative positional embedding used in HuBERT
    """

    def __init__(self, dim=768, kernel_size=128, groups=16):
        super().__init__()
        self.conv = nn.Conv1d(dim, dim, kernel_size=kernel_size, padding=kernel_size // 2, groups=groups, bias=True)
        self.conv = nn.utils.parametrizations.weight_norm(self.conv, name='weight', dim=2)

    def forward(self, x):
        x = self.conv(x)
        x = F.gelu(x[:, :, :-1])
        return x


INIT_ARGS = {
    'dim': 4,
    'kernel_size': 8,
    'groups': 2,
}

ARGS = {
    'x': torch.randn(2, 4, 16),
}

if __name__ == '__main__':
    model = PositionalConvEmbedding(**INIT_ARGS)
    output = model(**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 3, dtype family float, shape (2, 4, 16) — shape (2, 4, 16) 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)
RotaryEmbedding Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("4edb80e5146622da")
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 5, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

INIT_ARGS = {
    'dim': 8
}

ARGS = {
    'q': torch.randn(2, 4, 3, 8),
    'k': torch.randn(2, 4, 3, 8)
}

def rotate_half(x):
    x1, x2 = x.chunk(2, dim=-1)
    return torch.cat((-x2, x1), dim=-1)

def apply_rotary_pos_emb(x, cos, sin):
    cos = cos[:, :, :x.shape[-2], :]
    sin = sin[:, :, :x.shape[-2], :]
    return x * cos + rotate_half(x) * sin

class RotaryEmbedding(torch.nn.Module):
    def __init__(self, dim: int):
        super().__init__()
        inv_freq = 1.0 / 10000 ** (torch.arange(0, dim, 2).float() / dim)
        self.register_buffer('inv_freq', inv_freq)
        self._seq_len_cached = None
        self._cos_cached = None
        self._sin_cached = None

    def _update_cos_sin_tables(self, x, seq_dimension=-2):
        seq_len = x.shape[seq_dimension]
        if seq_len != self._seq_len_cached or self._cos_cached.device != x.device or self._cos_cached.dtype != x.dtype:
            self._seq_len_cached = seq_len
            t = torch.arange(x.shape[seq_dimension], device=x.device, dtype=torch.float32)
            freqs = torch.einsum('i,j->ij', t, self.inv_freq.to(x.dtype))
            emb = torch.cat((freqs, freqs), dim=-1).to(x.device)
            self._cos_cached = emb.cos()[None, None, :, :].to(x.dtype)
            self._sin_cached = emb.sin()[None, None, :, :].to(x.dtype)
        return (self._cos_cached, self._sin_cached)

    def forward(self, q, k):
        self._cos_cached, self._sin_cached = self._update_cos_sin_tables(q.float(), seq_dimension=-2)
        if k is not None:
            return (apply_rotary_pos_emb(q.float(), self._cos_cached, self._sin_cached).type_as(q), apply_rotary_pos_emb(k.float(), self._cos_cached, self._sin_cached).type_as(k))
        else:
            return (apply_rotary_pos_emb(q.float(), self._cos_cached, self._sin_cached).type_as(q), None)

if __name__ == '__main__':
    model = RotaryEmbedding(**INIT_ARGS)
    output = model(**ARGS)
    print(f"Output shapes: q={output[0].shape}, k={output[1].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, ndim 5, 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)
SinusoidalPositionalEncoding Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("64b7554453a70d14")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 8, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import math

class SinusoidalPositionalEncoding(nn.Module):

    def __init__(self, dim, length):
        super(SinusoidalPositionalEncoding, self).__init__()
        self.length = length
        self.dim = dim
        self.register_buffer('pe', self._generate_positional_encoding(length, dim))

    def _generate_positional_encoding(self, length, dim):
        pe = torch.zeros(length, dim)
        position = torch.arange(0, length, dtype=torch.float).unsqueeze(1)
        div_term = torch.exp(torch.arange(0, dim, 2).float() * (-math.log(10000.0) / dim))
        pe[:, 0::2] = torch.sin(position * div_term)
        pe[:, 1::2] = torch.cos(position * div_term)
        pe = pe.unsqueeze(0)
        return pe

    def forward(self, x):
        x = x + self.pe[:, :x.size(1)]
        return x


INIT_ARGS = {
    'dim': 4,
    'length': 8,
}

ARGS = {
    'x': torch.randn(2, 8, 4),
}

if __name__ == '__main__':
    model = SinusoidalPositionalEncoding(**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 3, dtype family float, shape (2, 8, 4) — shape (2, 8, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
Snake Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("1ba59f2c2995a11d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (1, 2, 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.nn as nn

def snake_beta(x, alpha, beta):
    return x + beta * torch.sin(x * alpha).pow(2)

class Snake(nn.Module):

    def __init__(self, dim_in, dim_out, bias, alpha_trainable=True):
        super().__init__()
        self.proj = nn.Linear(dim_in, dim_out, bias=bias)
        self.alpha = nn.Parameter(torch.ones(1, 1, dim_out))
        self.beta = nn.Parameter(torch.ones(1, 1, dim_out))
        self.alpha.requires_grad = alpha_trainable
        self.beta.requires_grad = alpha_trainable

    def forward(self, x):
        x = self.proj(x)
        x = snake_beta(x, self.alpha, self.beta)
        return x


INIT_ARGS = {
    'dim_in': 4,
    'dim_out': 3,
    'bias': True,
    'alpha_trainable': True,
}

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

if __name__ == '__main__':
    snake = Snake(**INIT_ARGS)
    output = snake(**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 3, dtype family float, shape (1, 2, 3) — shape (1, 2, 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)
TimestepEmbedder Ran
ran on a synthesized input
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("ec75a1f98a2e71d6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (2, 12) 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 nn
import math

def timestep_embedding(timesteps, dim, max_period=10000):
    """
    Create sinusoidal timestep embeddings.

    :param timesteps: a 1-D Tensor of N indices, one per batch element.
                      These may be fractional.
    :param dim: the dimension of the output.
    :param max_period: controls the minimum frequency of the embeddings.
    :return: an [N x dim] Tensor of positional embeddings.
    """
    half = dim // 2
    freqs = torch.exp(-math.log(max_period) * torch.arange(start=0, end=half, dtype=torch.float32) / half).to(device=timesteps.device)
    args = timesteps[:, None].float() * freqs[None]
    embedding = torch.cat([torch.cos(args), torch.sin(args)], dim=-1)
    if dim % 2:
        embedding = torch.cat([embedding, torch.zeros_like(embedding[:, :1])], dim=-1)
    return embedding

class TimestepEmbedder(nn.Module):
    """
    Embeds scalar timesteps into vector representations.
    """

    def __init__(self, hidden_size, frequency_embedding_size=256, out_size=None):
        super().__init__()
        if out_size is None:
            out_size = hidden_size
        self.mlp = nn.Sequential(nn.Linear(frequency_embedding_size, hidden_size, bias=True), nn.SiLU(), nn.Linear(hidden_size, out_size, bias=True))
        self.frequency_embedding_size = frequency_embedding_size

    def forward(self, t):
        t_freq = timestep_embedding(t, self.frequency_embedding_size).type(self.mlp[0].weight.dtype)
        t_emb = self.mlp(t_freq)
        return t_emb


INIT_ARGS = {
    'hidden_size': 8,
    'frequency_embedding_size': 16,
    'out_size': 12
}

ARGS = {
    't': torch.tensor([0, 1], dtype=torch.long)
}

if __name__ == '__main__':
    model = TimestepEmbedder(**INIT_ARGS)
    output = model(**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 2, dtype family float, shape (2, 12) — shape (2, 12) 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)
apply_rotary_pos_emb Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("206c90f5d5d477d1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 4, dtype family float, shape (2, 3, 4, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create input tensor x with shape (batch=2, heads=3, seq_len=4, head_dim=4)
# The function chunks on dim=-1, so head_dim must be even
x = torch.randn(2, 3, 4, 4)

# Create cos tensor with shape (batch=2, heads=3, max_seq_len=5, head_dim=4)
# The function slices [:, :, :x.shape[-2], :] so we need seq_len dimension > x's seq_len
cos = torch.randn(2, 3, 5, 4)

# Create sin tensor with same shape as cos
sin = torch.randn(2, 3, 5, 4)

ARGS = {
    "x": x,
    "cos": cos,
    "sin": sin,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 4, 4) — shape (2, 3, 4, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_mask_indices Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("8b8f03240223d1ab")
How it ran

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

import torch
import numpy as np

torch.manual_seed(0)

# Create small tensors for testing
batch_size = 2
timesteps = 4

# Optional padding mask - shape (batch_size, timesteps)
# True values indicate padding, False indicates valid positions
padding_mask = torch.tensor([
    [False, False, True, True],
    [False, False, False, True]
])

ARGS = {
    'shape': (batch_size, timesteps),
    'padding_mask': padding_mask,
    'mask_prob': 0.5,
    'mask_length': 2,
    'mask_type': 'static',
    'mask_other': 0.1,
    'min_masks': 1,
    'no_overlap': False,
    'min_space': 0
}
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 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
create_mask Ran
ran; our contract draft was wrong, not the code
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("2004128767739013")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 4, dtype family float, shape (2, 1, 4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from einops import rearrange, repeat

torch.manual_seed(0)

# Analyze the function signature and body:
# - q_shape: tuple used as q_shape[0] (batch), q_shape[-2] (sequence length i)
# - k_shape: tuple used as k_shape[-2] (sequence length j)
# - device: string or torch.device
# - q_mask: optional tensor, defaults to ones((b, i), device=device, dtype=torch.bool)
# - k_mask: optional tensor, defaults to ones((b, j), device=device, dtype=torch.bool)

# Create small tensors with distinct dimensions
batch_size = 2
q_seq_len = 4
k_seq_len = 3

device = 'cpu'

# q_shape should have batch at index 0 and sequence length at index -2
# Common pattern: (batch, heads, seq_len, dim) or (batch, seq_len, dim)
q_shape = (batch_size, 8, q_seq_len, 64)  # (batch, heads, seq_len, dim)
k_shape = (batch_size, 8, k_seq_len, 64)  # (batch, heads, seq_len, dim)

# Optional masks - can be None to test defaults, or provide explicit ones
q_mask = torch.ones((batch_size, q_seq_len), device=device, dtype=torch.bool)
k_mask = torch.ones((batch_size, k_seq_len), device=device, dtype=torch.bool)

ARGS = {
    'q_shape': q_shape,
    'k_shape': k_shape,
    'device': device,
    'q_mask': q_mask,
    'k_mask': k_mask,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 1, 4, 3) — shape (2, 1, 4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
film_modulate Ran
ran, honoured the contract we drafted
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("0e9be0fe6b203878")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# Construct tiny tensors for the function arguments
# film_modulate performs: x * (1 + scale) + shift
# All three parameters are used with arithmetic operations

x = torch.randn(2, 4)  # batch size 2, dimension 4
shift = torch.randn(2, 4)  # same shape as x
scale = torch.randn(2, 4)  # same shape as x

ARGS = {
    "x": x,
    "shift": shift,
    "scale": scale,
}
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
snake_beta Ran
ran; our contract draft was wrong, not the code
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("8f8b4880e8072344")
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)

# Construct arguments for snake_beta function
x = torch.randn(2, 4)  # Small tensor: batch size 2, dimension 4
alpha = 0.5  # Scalar parameter for modulation
beta = 0.3   # Scalar parameter for amplitude

ARGS = {
    "x": x,
    "alpha": alpha,
    "beta": beta
}
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
Attention Not yet run
unverified (harvested, no recorded run)
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("531168589467d2a8")
How it ran

no invocation recorded.

DiTBlock Not yet run
unverified (harvested, no recorded run)
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("d23e4f9c8defff20")
How it ran

no invocation recorded.

DiTControlNet Not yet run
unverified (harvested, no recorded run)
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("7f5058304c41fd44")
How it ran

no invocation recorded.

DiTControlNetEmbed Not yet run
unverified (harvested, no recorded run)
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("56037e5ee5cc1c1f")
How it ran

no invocation recorded.

add_mask Not yet run
unverified (harvested, no recorded run)
haidog-yaqub/EzAudio/src/models/controlnet.py
code served (permissive licence) · get_code("aefbf29ffedaf4a6")
How it ran

no invocation recorded.

Also present in other papers' code

Of 23 samples lifted from this paper's linked repositories, 23 carry a function identity and 0 do not. Of the 23 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 22 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
apply_rotary_pos_emb
brick of this paper's assembly: helper
2 other papers
2 dated earlier
2406.10391 (2024-06)
BEACON: Benchmark for Comprehensive RNA Tasks and Language Models
2410.22949 (2024-10)
MutaPLM: Protein Language Modeling for Mutation Explanation and Engineering

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

Repositories linked to this paper

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

Abstract

We study the fine-grained text-to-audio (T2A) generation task. While recent models can synthesize high-quality audio from text descriptions, they often lack precise control over attributes such as loudness, pitch, and sound events. Unlike prior approaches that retrain models for specific control types, we propose to train ControlNet models on top of pre-trained T2A backbones to achieve controllable generation over loudness, pitch, and event roll. We introduce two designs, T2A-ControlNet and T2A-Adapter, and show that the T2A-Adapter model offers a more efficient structure with strong control ability. With only 38M additional parameters, T2A-Adapter achieves stateof-the-art performance on the AudioSet-Strong in both event-level and segment-level F1 scores. We further extend this framework to audio editing, proposing T2A-Editor for removing and inserting audio events at time locations specified by instructions. Models, code, dataset pipelines, and benchmarks will be released to support future research on controllable audio generation and editing.

For agents

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

get_harvested_code_for_paper("2602.04680")
get_code_for_paper("2602.04680")
have("2602.04680")

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

curl https://syntology.ai/api/ran/2602.04680.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.04680.svg)](https://syntology.ai/paper/2602.04680)

Connect an agent — have() is free.