SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2011.12913 · NeurIPS · 2020

torchdistill: A Modular, Configuration-Driven Framework for Knowledge Distillation

Yoshitomo Matsubara

arXiv · PDF · Open in the Atlas

Code that ran

10 of 26 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 10 failures of our instrument; 16 no recorded run), measured 2026-11-21; 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/2011.12913.json · badge: /api/ran/2011.12913.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
yoshitomo-matsubara/torchdistill canonical (classifier label)
classifier basis: classifier_2025
10 ran · 16 no recorded run
FunctionStatusWhere it lives
_resolve_module_path_flag Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("c3ed65ba6ea90613")
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 set. 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)

# Test case 1: flag_config is None
flag_config_1 = None
target_module_path_set_1 = {"layer1", "layer2", "layer3"}

# Test case 2: flag_config is False
flag_config_2 = False
target_module_path_set_2 = {"layer1", "layer2", "layer3"}

# Test case 3: flag_config is True
flag_config_3 = True
target_module_path_set_3 = {"layer1", "layer2", "layer3"}

# Test case 4: flag_config is a list of valid module paths
flag_config_4 = ["layer1", "layer2"]
target_module_path_set_4 = {"layer1", "layer2", "layer3"}

# Using the most complete test case for ARGS
ARGS = {
    "flag_config": ["layer1", "layer2"],
    "target_module_path_set": {"layer1", "layer2", "layer3"}
}
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 set — a set output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a set 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
build_proc_func Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("3254f2af805330f9")
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 torch.nn as nn
from functools import wraps

torch.manual_seed(0)

# Create a simple registered process function registry
registered_funcs = {}

def simple_proc_func(self, x, multiplier=1, offset=0):
    """A simple process function for testing"""
    return x * multiplier + offset

# Register the function
registered_funcs['simple_processor'] = simple_proc_func

def get_registered_proc_func(key):
    """Mock getter for registered process functions"""
    if key not in registered_funcs:
        raise ValueError(f"Process function '{key}' not registered")
    return registered_funcs[key]

# Test case 1: proc_config as string
proc_config_str = 'simple_processor'

# Test case 2: proc_config as dict without kwargs
proc_config_dict_no_kwargs = {
    'key': 'simple_processor'
}

# Test case 3: proc_config as dict with kwargs
proc_config_dict_with_kwargs = {
    'key': 'simple_processor',
    'kwargs': {'multiplier': 2, 'offset': 5}
}

ARGS = {
    'proc_config': proc_config_dict_with_kwargs,
    'get_registered_proc_func': get_registered_proc_func
}
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
get_device_index Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("87b46952aaf35391")
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 str. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
from collections import abc

torch.manual_seed(0)

# Test case 1: Single tensor on CPU
tensor_cpu = torch.randn(2, 4)

# Test case 2: Dictionary containing a tensor
dict_with_tensor = {
    'data': torch.randn(2, 4),
    'label': 'some_label'
}

# Test case 3: Tuple containing a tensor
tuple_with_tensor = (torch.randn(2, 4), 'metadata')

# Test case 4: Nested structure with no tensors
nested_no_tensor = {
    'values': [1, 2, 3],
    'info': ('a', 'b')
}

# Test case 5: List (which is a Sequence but handled specially)
list_with_tensor = [torch.randn(2, 4), torch.randn(3, 5)]

# Test case 6: Custom sequence (not list or tuple)
class CustomSequence(abc.Sequence):
    def __init__(self, items):
        self.items = items
    
    def __getitem__(self, idx):
        return self.items[idx]
    
    def __len__(self):
        return len(self.items)

custom_seq = CustomSequence([torch.randn(2, 4), 'data'])

# Use the simplest case for ARGS
ARGS = {
    'data': tensor_cpu
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
get_proc_config Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("fd6d7563a4a31ea5")
How it ran

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

import warnings
import logging

torch_seed = 0

# Set up logger
def_logger = logging.getLogger()
logger = def_logger.getChild(__name__)

# Test case 1: key exists in train_config
train_config_1 = {
    'pre_forward_proc': {'enabled': True, 'type': 'custom'}
}

# Test case 2: deprecated key exists in train_config
train_config_2 = {
    'pre_forward_process': {'enabled': False, 'type': 'legacy'}
}

# Test case 3: neither key exists
train_config_3 = {
    'some_other_key': 'value'
}

# Use the first test case as the primary one
ARGS = {
    'train_config': train_config_1,
    'key': 'pre_forward_proc'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
register_forward_hook_with_dict Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("4b491e5e58b0f881")
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 RemovableHandle. 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)

class SimpleModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

class RootModule(nn.Module):
    def __init__(self):
        super().__init__()
        self.child = SimpleModule()
    
    def forward(self, x):
        return self.child(x)

root_module = RootModule()
module_path = "child"
requires_input = True
requires_output = True
io_dict = {}
accumulates = False

ARGS = {
    'root_module': root_module,
    'module_path': module_path,
    'requires_input': requires_input,
    'requires_output': requires_output,
    'io_dict': io_dict,
    'accumulates': accumulates
}
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 RemovableHandle — a RemovableHandle output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a RemovableHandle 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
register_post_epoch_proc_func Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("c92bc6d24855c49a")
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 torch.nn as nn

torch.manual_seed(0)

# Define a simple post-epoch process function for testing
def my_post_epoch_func(self, metrics=None, **kwargs):
    """A simple post-epoch process function."""
    pass

# Alternative function with a different name
def another_post_epoch_func(self, metrics=None, **kwargs):
    """Another post-epoch process function."""
    pass

# Test case 1: Using the decorator with a callable directly
ARGS = {
    'arg': my_post_epoch_func,
    'kwargs': {}
}
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
register_post_forward_proc_func Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("7c61d41d0c3e500d")
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 torch.nn as nn

torch.manual_seed(0)

# Define a simple post-forward process function for testing
def my_post_forward_proc(self, loss, metrics=None, **kwargs):
    """A simple post-forward process function."""
    return loss

# Define another function to test the decorator
def another_post_forward_proc(self, loss, metrics=None, **kwargs):
    """Another post-forward process function."""
    return loss

# Create test arguments for the register_post_forward_proc_func function
ARGS = {
    'arg': my_post_forward_proc,
    'kwargs': {}
}

# Alternative test case with a key parameter
ARGS_WITH_KEY = {
    'arg': another_post_forward_proc,
    'kwargs': {'key': 'custom_post_forward_proc'}
}

# Test case with arg=None (using decorator pattern)
ARGS_DECORATOR = {
    'arg': None,
    'kwargs': {'key': 'my_custom_key'}
}

# Use the first set of arguments as the primary ARGS
ARGS = 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 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
register_pre_epoch_proc_func Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("b354561e7df46cce")
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 torch.nn as nn

torch.manual_seed(0)

# Define a simple pre-epoch process function to use as test argument
def sample_pre_epoch_proc(self, epoch=None, **kwargs):
    """Sample pre-epoch process function"""
    pass

def another_pre_epoch_proc(self, epoch=None, **kwargs):
    """Another sample pre-epoch process function"""
    pass

# Test case 1: Calling with a function as positional argument
ARGS = {
    'arg': sample_pre_epoch_proc,
    'kwargs': {}
}

# Alternative test case 2: Calling with decorator pattern and key
ARGS_WITH_KEY = {
    'arg': another_pre_epoch_proc,
    'kwargs': {'key': 'custom_pre_epoch_func'}
}

# Use the first variant as the main ARGS
ARGS = {
    'arg': sample_pre_epoch_proc,
    'kwargs': {}
}
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
register_pre_forward_proc_func Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("6a583218d3dff95c")
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 torch.nn as nn

torch.manual_seed(0)

# Define a simple pre-forward process function for testing
def sample_pre_forward_proc(self, *args, **kwargs):
    """Sample pre-forward process function"""
    pass

def another_pre_forward_proc(self, *args, **kwargs):
    """Another sample pre-forward process function"""
    pass

# ARGS dict with all parameters for register_pre_forward_proc_func
ARGS = {
    'arg': sample_pre_forward_proc,
}
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
set_hooks Ran
ran; our contract draft was wrong, not the code
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("f44f1f4a4378af5a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned list, ndim 2, dtype family 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
from unittest.mock import Mock, MagicMock

torch.manual_seed(0)

# Create a simple model with named submodules
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.layer1 = nn.Linear(4, 3)
        self.layer2 = nn.Linear(3, 2)
    
    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        return x

# Create model instances
model = SimpleModel()
unwrapped_org_model = SimpleModel()

# Create model_config with forward_hook configuration
model_config = {
    'forward_hook': {
        'input': ['layer1'],
        'output': ['layer2'],
        'accumulates': False,
        'stacks_accumulated': False
    }
}

# Create a mock ForwardHookManager
class MockForwardHookManager:
    def __init__(self):
        self.hooks = []
    
    def add_hook_to_module(self, module, module_path, requires_input=False, requires_output=False, accumulates=False, stacks_accumulated=False):
        # Mock handle
        mock_handle = Mock()
        mock_handle.remove = Mock()
        pair = (module_path, mock_handle)
        self.hooks.append(pair)
        return pair

forward_hook_manager = MockForwardHookManager()

# Stub the get_module function that is used internally
def get_module(model, module_path):
    parts = module_path.split('.')
    module = model
    for part in parts:
        module = getattr(module, part)
    return module

# Inject get_module into the module scope for the function to use
import sys
current_module = sys.modules[__name__]
current_module.get_module = get_module

ARGS = {
    'model': model,
    'unwrapped_org_model': unwrapped_org_model,
    'model_config': model_config,
    'forward_hook_manager': forward_hook_manager
}
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 list, ndim 2, dtype family float — a list output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
DistillationBox Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("5e932d9e557d1152")
How it ran

no invocation recorded.

ForwardHookManager Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("733dc5424bd2b106")
How it ran

no invocation recorded.

_extract_module Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("62cb0c77042128ad")
How it ran

no invocation recorded.

clear_io_dict Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("3f67b3fedf426f90")
How it ran

no invocation recorded.

clear_io_dict_values Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("0b996e294219e230")
How it ran

no invocation recorded.

default_post_epoch_process_with_teacher Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("7b5cddaa746c727a")
How it ran

no invocation recorded.

default_post_forward_process Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("25556f7f6770ba26")
How it ran

no invocation recorded.

default_pre_epoch_process_with_teacher Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("4d443b9aef2436a9")
How it ran

no invocation recorded.

default_pre_forward_process Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("aaa26d595f56f023")
How it ran

no invocation recorded.

get_forward_proc_func Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("d6b164151ea048bc")
How it ran

no invocation recorded.

get_post_epoch_proc_func Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("c2780237c1682ad8")
How it ran

no invocation recorded.

get_post_forward_proc_func Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("e78710c20c16f470")
How it ran

no invocation recorded.

get_pre_epoch_proc_func Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("d75c1131fcde7500")
How it ran

no invocation recorded.

get_pre_forward_proc_func Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("7f13327e06b694aa")
How it ran

no invocation recorded.

update_io_dict Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("3933a969d27c0f3a")
How it ran

no invocation recorded.

wrap_model Not yet run
unverified (harvested, no recorded run)
yoshitomo-matsubara/torchdistill/torchdistill/core/distillation.py
code served (permissive licence) · get_code("0c2e416b9fd8f9e7")
How it ran

no invocation recorded.

Also present in other papers' code

Of 26 samples lifted from this paper's linked repositories, 26 carry a function identity and 0 do not. Of the 26 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 26 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.

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 2026-11-21.

Repositories linked to this paper

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

Abstract

While knowledge distillation (transfer) has been attracting attentions from the research community, the recent development in the fields has heightened the need for reproducible studies and highly generalized frameworks to lower barriers to such high-quality, reproducible deep learning research. Several researchers voluntarily published frameworks used in their knowledge distillation studies to help other interested researchers reproduce their original work. Such frameworks, however, are usually neither well generalized nor maintained, thus researchers are still required to write a lot of code to refactor/build on the frameworks for introducing new methods, models, datasets and designing experiments. In this paper, we present our developed open-source framework built on PyTorch and dedicated for knowledge distillation studies. The framework is designed to enable users to design experiments by declarative PyYAML configuration files, and helps researchers complete the recently proposed ML Code Completeness Checklist. Using the developed framework, we demonstrate its various efficient training strategies, and implement a variety of knowledge distillation methods. We also reproduce some of their original experimental results on the ImageNet and COCO datasets presented at major machine learning conferences such as ICLR, NeurIPS, CVPR and ECCV, including recent state-of-the-art methods. All the source code, configurations, log files and trained model weights are publicly available at https://github.com/yoshitomo-matsubara/torchdistill.

For agents

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

get_harvested_code_for_paper("2011.12913")
get_code_for_paper("2011.12913")
have("2011.12913")

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

curl https://syntology.ai/api/ran/2011.12913.json

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

[![Syntology run record](https://syntology.ai/api/ran/2011.12913.svg)](https://syntology.ai/paper/2011.12913)

Connect an agent — have() is free.