SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2602.15537 · 2026

ZeroSyl: Simple Zero-Resource Syllable Tokenization for Spoken Language Modeling

Herman Kamper, Danel Slabbert, Simon Malan, Nicol Visser

arXiv · PDF · Open in the Atlas

Code that ran

14 of 20 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (8 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 9 have no contract verdict; 5 failures of our instrument; 6 no recorded run), measured 2026-11-22; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2602.15537.json · badge: /api/ran/2602.15537.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
nicolvisser/ZeroSyl community
Papers with Code archive
14 ran · 6 no recorded run
FunctionStatusWhere it lives
ConvFeatureExtractionModel Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("c9f97dfc6d4d3717")
How it ran

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

import torch
import torch.nn as nn
import math

torch.manual_seed(0)

# Define the ARGS dictionary with all required parameters for ConvFeatureExtractionModel
ARGS = {
    "conv_layers": [
        (64, 10, 5),
        (128, 3, 2),
        (256, 3, 2),
    ],
    "dropout": 0.1,
    "mode": "default",
    "conv_bias": False,
    "conv_type": "default",
}

# Verify the model can be instantiated and run forward pass
if __name__ == "__main__":
    from torch.nn.modules.module import Module
    
    # Import the class (assuming it's defined above)
    model = nn.Module()  # Placeholder - in real execution the class would be imported
    
    # Create a small test input
    batch_size = 2
    seq_length = 160
    x = torch.randn(batch_size, seq_length)
    
    print("ARGS dictionary created successfully")
    print(f"conv_layers: {ARGS['conv_layers']}")
    print(f"dropout: {ARGS['dropout']}")
    print(f"mode: {ARGS['mode']}")
    print(f"conv_bias: {ARGS['conv_bias']}")
    print(f"conv_type: {ARGS['conv_type']}")
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 ConvFeatureExtractionModel — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ConvFeatureExtractionModel 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
Fp32GroupNorm Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("238d038637216da2")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Fp32GroupNorm: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Fp32GroupNorm inherits from nn.GroupNorm
# nn.GroupNorm.__init__ signature: (num_groups, num_channels, eps=1e-5, affine=True)
# The forward method uses: input, self.num_groups, self.weight, self.bias, self.eps

# Create a small Fp32GroupNorm instance with appropriate parameters
num_groups = 2
num_channels = 4
eps = 1e-5
affine = True

model = nn.GroupNorm(num_groups=num_groups, num_channels=num_channels, eps=eps, affine=affine)

# Create a small input tensor: batch_size=2, channels=4, spatial_dims=(2, 2)
# Shape should be (batch, channels, *spatial_dims) for group norm
input_tensor = torch.randn(2, 4, 2, 2)

ARGS = {
    'num_groups': num_groups,
    'num_channels': num_channels,
    'eps': eps,
    'affine': affine,
}
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 Fp32GroupNorm — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Fp32GroupNorm 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
Fp32LayerNorm Ran
ran on a synthesized input
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("bee6f7686b4cd186")
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 Fp32LayerNorm(nn.LayerNorm):

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)

    def forward(self, input):
        output = F.layer_norm(input.float(), self.normalized_shape, self.weight.float() if self.weight is not None else None, self.bias.float() if self.bias is not None else None, self.eps)
        return output.type_as(input)


INIT_ARGS = {
    'normalized_shape': (4,),
}

ARGS = {
    'input': torch.randn(2, 4, dtype=torch.float16),
}

if __name__ == '__main__':
    layer_norm = Fp32LayerNorm(**INIT_ARGS)
    output = layer_norm(**ARGS)
    print(f"Input shape: {ARGS['input'].shape}, dtype: {ARGS['input'].dtype}")
    print(f"Output shape: {output.shape}, dtype: {output.dtype}")
    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 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)
GLU_Linear Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("5ec66269b12c64b0")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'input_dim': 8,
    'output_dim': 4,
    'glu_type': 'sigmoid',
    'bias_in_glu': True
}
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 GLU_Linear — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a GLU_Linear 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
MultiheadAttention Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("11bbf68104c4afcd")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    'embed_dim': 8,
    'num_heads': 2,
    'kdim': 8,
    'vdim': 8,
    'dropout': 0.1,
    'bias': True,
    'add_bias_kv': False,
    'add_zero_attn': False,
    'self_attention': True,
    'encoder_decoder_attention': False,
    'q_noise': 0.0,
    'qn_block_size': 4,
    'has_relative_attention_bias': False,
    'num_buckets': 32,
    'max_distance': 128,
    'gru_rel_pos': False,
    'rescale_init': False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type MultiheadAttention — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a MultiheadAttention 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
SamePad Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("e6b0e1474771c0f8")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# The SamePad module takes kernel_size and causal as parameters
# In __init__, it uses kernel_size to compute self.remove based on whether causal is True
# In forward, it expects x to be a 3D tensor (batch, channels, length) and slices it along dim 2

# Create test arguments
ARGS = {
    "kernel_size": 3,
    "causal": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SamePad — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a SamePad 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
Swish Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("361f8f73bf615221")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# The Swish class only has __init__ and forward methods
# __init__ takes no parameters besides self
# forward takes x (a tensor) as input

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

ARGS = {}
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 Swish — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Swish 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
TransposeLast Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("02f2ef2ed6d05541")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# The TransposeLast module takes one parameter: deconstruct_idx
# which can be None or an integer index used for indexing into x before transposing

# Create a sample tensor to understand the expected input shape
# x should be at least 2D since we transpose the last two dimensions
# Using shape (2, 4, 3, 5) - batch=2, then other dimensions distinct for probing
sample_tensor = torch.randn(2, 4, 3, 5)

# Test case 1: deconstruct_idx=None (simplest case)
ARGS = {
    "deconstruct_idx": None
}
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 TransposeLast — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a TransposeLast 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
WavLMConfig Ran
ran on a synthesized input · constructed an object; no result was computed
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("47b9fcbf6c96bc4a")
How it ran

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

import torch

torch.manual_seed(0)

ARGS = {
    'cfg': {
        'extractor_mode': 'default',
        'encoder_layers': 12,
        'encoder_embed_dim': 768,
        'encoder_ffn_embed_dim': 3072,
        'encoder_attention_heads': 12,
        'activation_fn': 'gelu',
        'layer_norm_first': False,
        'conv_feature_layers': '[(512,10,5)] + [(512,3,2)] * 4 + [(512,2,2)] * 2',
        'conv_bias': False,
        'feature_grad_mult': 1.0,
        'normalize': False,
        'dropout': 0.1,
        'attention_dropout': 0.1,
        'activation_dropout': 0.0,
        'encoder_layerdrop': 0.0,
        'dropout_input': 0.0,
        'dropout_features': 0.0,
        'mask_length': 10,
        'mask_prob': 0.65,
        'mask_selection': 'static',
        'mask_other': 0,
        'no_mask_overlap': False,
        'mask_min_space': 1,
        'mask_channel_length': 10,
        'mask_channel_prob': 0.0,
        'mask_channel_selection': 'static',
        'mask_channel_other': 0,
        'no_mask_channel_overlap': False,
        'mask_channel_min_space': 1,
        'conv_pos': 128,
        'conv_pos_groups': 16,
        'relative_position_embedding': False,
        'num_buckets': 320,
        'max_distance': 1280,
        'gru_rel_pos': False,
    }
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type WavLMConfig — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a WavLMConfig output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
compute_mask_indices Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("147814c7719109eb")
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 ndarray, ndim 2, dtype family float, shape (2, 10) 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)
np.random.seed(0)

# Construct arguments for compute_mask_indices
shape = (2, 10)  # batch_size=2, timesteps=10

# Create optional padding mask (same shape as (batch_size, timesteps))
# True values indicate padding that should not be masked
padding_mask = torch.tensor([
    [False, False, False, False, False, True, True, True, True, True],
    [False, False, False, False, False, False, False, True, True, True]
])

mask_prob = 0.15
mask_length = 3
mask_type = 'static'
mask_other = 0.0
min_masks = 1
no_overlap = False
min_space = 0

ARGS = {
    'shape': shape,
    'padding_mask': padding_mask,
    'mask_prob': mask_prob,
    'mask_length': mask_length,
    'mask_type': mask_type,
    'mask_other': mask_other,
    'min_masks': min_masks,
    'no_overlap': no_overlap,
    'min_space': min_space,
}
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 ndarray, ndim 2, dtype family float, shape (2, 10) — shape (2, 10) 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
gelu Ran
ran; our contract draft was wrong, not the code
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("f30ebf4e48b7e2d7")
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)

# Create a small tensor with shape (2, 4)
x = torch.randn(2, 4, dtype=torch.float32)

ARGS = {
    "x": x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
gelu_accurate Ran
ran; our contract draft was wrong, not the code
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("581b5373c5b2c0a0")
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 math

torch.manual_seed(0)

# The function gelu_accurate takes a single parameter x
# Looking at the body, x is used with:
# - torch.tanh() - needs a tensor
# - torch.pow(x, 3) - needs a tensor
# - arithmetic operations (+, *, etc.)
# So x should be a torch.Tensor

x = torch.randn(2, 4)

ARGS = {
    'x': x
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_activation_fn Ran
ran; our contract draft was wrong, not the code
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("591bd7145f129c23")
How it ran

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

import torch
import math
import torch.nn.functional as F
import warnings

torch.manual_seed(0)

ARGS = {
    'activation': 'gelu'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
quant_noise Ran
ran; our contract draft was wrong, not the code
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("50d074f5c3f3407d")
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 Linear. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple Linear module for testing
linear_module = nn.Linear(in_features=8, out_features=4)

# Create a simple Embedding module for testing
embedding_module = nn.Embedding(num_embeddings=10, embedding_dim=8)

# Create a simple Conv2d module with 1x1 kernel
conv_1x1_module = nn.Conv2d(in_channels=8, out_channels=4, kernel_size=1)

# Create a simple Conv2d module with larger kernel
conv_3x3_module = nn.Conv2d(in_channels=8, out_channels=4, kernel_size=3, padding=1)

ARGS = {
    "module": linear_module,
    "p": 0.5,
    "block_size": 2,
}
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 Linear — a Linear output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Linear 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
TransformerEncoder Not yet run
unverified (harvested, no recorded run)
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("ad5e79e9366f6b46")
How it ran

no invocation recorded.

TransformerSentenceEncoderLayer Not yet run
unverified (harvested, no recorded run)
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("8e00a2f405df6e94")
How it ran

no invocation recorded.

WavLM Not yet run
unverified (harvested, no recorded run)
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("b784924ce3753e6a")
How it ran

no invocation recorded.

ZeroSylContinuous Not yet run
unverified (harvested, no recorded run)
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("da73d28df9e96f2d")
How it ran

no invocation recorded.

ZeroSylDiscrete Not yet run
unverified (harvested, no recorded run)
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("0a1ed7e7ae39261a")
How it ran

no invocation recorded.

init_bert_params Not yet run
unverified (harvested, no recorded run)
nicolvisser/ZeroSyl/zerosyl/zerosyl.py
code served (permissive licence) · get_code("f125b0d1fe596f71")
How it ran

no invocation recorded.

Also present in other papers' code

Of 20 samples lifted from this paper's linked repositories, 20 carry a function identity and 0 do not. Of the 20 distinct functions with an identity, 10 are also present in other papers whose code we hold (5 with at least one paper dated earlier, 4 the same month, 6 later) and 10 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
gelu
brick of this paper's assembly: helper
11 other papers
9 dated earlier, 2 the same month
2204.02152 (2022-04)
UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
2204.03339 (2022-04)
Boosting Self-Supervised Embeddings for Speech Enhancement
2210.17016 (2022-10)
Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit
and 8 more not named here
gelu_accurate
brick of this paper's assembly: helper
11 other papers
9 dated earlier, 2 the same month
2204.02152 (2022-04)
UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
2204.03339 (2022-04)
Boosting Self-Supervised Embeddings for Speech Enhancement
2210.17016 (2022-10)
Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit
and 8 more not named here
get_activation_fn
brick of this paper's assembly: helper
11 other papers
9 dated earlier, 2 the same month
2204.02152 (2022-04)
UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
2204.03339 (2022-04)
Boosting Self-Supervised Embeddings for Speech Enhancement
2210.17016 (2022-10)
Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit
and 8 more not named here
compute_mask_indices
brick of this paper's assembly: helper
5 other papers
4 dated earlier, 1 the same month
2204.02152 (2022-04)
UTMOS: UTokyo-SaruLab System for VoiceMOS Challenge 2022
2204.03339 (2022-04)
Boosting Self-Supervised Embeddings for Speech Enhancement
2210.17016 (2022-10)
Wespeaker: A Research and Production oriented Speaker Embedding Learning Toolkit
and 2 more not named here
quant_noise
brick of this paper's assembly: helper
4 other papers
3 dated earlier, 1 later · 1 only as an assembly brick
2203.15610 (2022-03)
LightHuBERT: Lightweight and Configurable Speech Representation Learning with Once-for-All Hidden-Unit BERT
2307.13424 (2023-07)
Holistic Exploration on Universal Decompositional Semantic Parsing: Architecture, Data Augmentation, and LLM Paradigm
2407.03575 (2024-07)
DGR-MIL: Exploring Diverse Global Representation in Multiple Instance Learning for Whole Slide Image Classification
and 1 more not named here
Fp32GroupNorm
brick of this paper's assembly: helper
1 other paper
1 later
2606.17967 (2026-06)
Learning task-specific subspaces via interventional post-training of speech foundation models
Fp32LayerNorm
brick of this paper's assembly: helper
1 other paper
1 later
2606.17967 (2026-06)
Learning task-specific subspaces via interventional post-training of speech foundation models
SamePad
brick of this paper's assembly: helper
1 other paper
1 later
2606.17967 (2026-06)
Learning task-specific subspaces via interventional post-training of speech foundation models
TransposeLast
brick of this paper's assembly: helper
1 other paper
1 later
2606.17967 (2026-06)
Learning task-specific subspaces via interventional post-training of speech foundation models
init_bert_params
brick of this paper's assembly: helper
1 other paper
1 later
2606.17967 (2026-06)
Learning task-specific subspaces via interventional post-training of speech foundation models

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

Pure speech language models aim to learn language directly from raw audio without textual resources. A key challenge is that discrete tokens from self-supervised speech encoders result in excessively long sequences, motivating recent work on syllablelike units. However, methods like Sylber and SyllableLM rely on intricate multi-stage training pipelines. We propose Zero-Syl, a simple training-free method to extract syllable boundaries and embeddings directly from a frozen WavLM model. Using L2 norms of features in WavLM's intermediate layers, ZeroSyl achieves competitive syllable segmentation performance. The resulting segments are mean-pooled, discretized using K-means, and used to train a language model. ZeroSyl outperforms prior syllabic tokenizers across lexical, syntactic, and narrative benchmarks. Scaling experiments show that while finer-grained units are beneficial for lexical tasks, our discovered syllabic units exhibit better scaling behavior for syntactic modeling.

For agents

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

get_harvested_code_for_paper("2602.15537")
get_code_for_paper("2602.15537")
have("2602.15537")

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

curl https://syntology.ai/api/ran/2602.15537.json

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

[![Syntology run record](https://syntology.ai/api/ran/2602.15537.svg)](https://syntology.ai/paper/2602.15537)

Connect an agent — have() is free.