clean repo

fasterai/core/criteria.py

@@ -1,12 +1,12 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/core/00b_core.criteria.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/core/criteria.ipynb.
 
 # %% auto 0
 __all__ = ['random', 'large_final', 'squared_final', 'small_final', 'large_init', 'small_init', 'large_init_large_final',
            'small_init_small_final', 'magnitude_increase', 'movement', 'updating_magnitude_increase',
            'updating_movement', 'movmag', 'updating_movmag', 'criterias', 'Criteria', 'available_criterias',
            'grad_crit']
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 2
+# %% ../../nbs/core/criteria.ipynb 2
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -15,7 +15,7 @@
 from .granularity import *
 from typing import Callable
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 6
+# %% ../../nbs/core/criteria.ipynb 6
 class Criteria():
     def __init__(self, f:Callable, reducer:str='mean', normalizer:str=None, needs_init:bool=False, needs_update:bool=False, output_f:Callable=None, return_init=False):
         store_attr()
@@ -81,54 +81,54 @@ def update_weights(self, m):
         if self.needs_update: 
             m._old_weights = m.weight.data.clone() # The current value becomes the old one for the next iteration
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 9
+# %% ../../nbs/core/criteria.ipynb 9
 random = Criteria(torch.randn_like)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 12
+# %% ../../nbs/core/criteria.ipynb 12
 large_final = Criteria(torch.abs)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 15
+# %% ../../nbs/core/criteria.ipynb 15
 squared_final = Criteria(torch.square)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 18
+# %% ../../nbs/core/criteria.ipynb 18
 small_final = Criteria(compose(torch.abs, torch.neg))
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 21
+# %% ../../nbs/core/criteria.ipynb 21
 large_init = Criteria(torch.abs, needs_init=True, return_init=True)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 24
+# %% ../../nbs/core/criteria.ipynb 24
 small_init = Criteria(compose(torch.abs, torch.neg), needs_init=True, return_init=True)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 27
+# %% ../../nbs/core/criteria.ipynb 27
 large_init_large_final = Criteria(torch.abs, needs_init=True, output_f=torch.min)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 30
+# %% ../../nbs/core/criteria.ipynb 30
 small_init_small_final = Criteria(torch.abs, needs_init=True, output_f=lambda x,y: torch.neg(torch.max(x,y)))
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 33
+# %% ../../nbs/core/criteria.ipynb 33
 magnitude_increase = Criteria(torch.abs, needs_init=True, output_f= torch.sub)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 36
+# %% ../../nbs/core/criteria.ipynb 36
 movement = Criteria(noop, needs_init=True, output_f= lambda x,y: torch.abs(torch.sub(x,y)))
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 41
+# %% ../../nbs/core/criteria.ipynb 41
 updating_magnitude_increase = Criteria(torch.abs, needs_update=True, output_f= lambda x,y: torch.sub(x,y))
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 44
+# %% ../../nbs/core/criteria.ipynb 44
 updating_movement = Criteria(noop, needs_update=True, output_f= lambda x,y: torch.abs(torch.sub(x,y)))
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 47
+# %% ../../nbs/core/criteria.ipynb 47
 movmag = Criteria(noop, needs_init=True, output_f=lambda x,y: torch.abs(torch.mul(x, torch.sub(x,y))))
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 50
+# %% ../../nbs/core/criteria.ipynb 50
 updating_movmag = Criteria(noop, needs_update=True, output_f=lambda x,y: torch.abs(torch.mul(x, torch.sub(x,y))))
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 52
+# %% ../../nbs/core/criteria.ipynb 52
 criterias = ('random', 'large_final', 'small_final', 'squared_final', 'small_init', 'small_final', 'large_init_large_final', 'small_init_small_final', 'magnitude_increase', 'movement', 'updating_magnitude_increase', 'updating_movement', 'updating_movmag')
 def available_criterias():
     print(criterias)
 
-# %% ../../nbs/core/00b_core.criteria.ipynb 73
+# %% ../../nbs/core/criteria.ipynb 73
 def grad_crit(m, g):
     if g in granularities[m.__class__.__name__]: 
         dim = granularities[m.__class__.__name__][g]

fasterai/core/granularity.py

@@ -1,16 +1,16 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/core/00a_core.granularity.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/core/granularity.ipynb.
 
 # %% auto 0
 __all__ = ['Granularities']
 
-# %% ../../nbs/core/00a_core.granularity.ipynb 2
+# %% ../../nbs/core/granularity.ipynb 2
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 from fastcore.basics import *
 from fastcore.imports import *
 
-# %% ../../nbs/core/00a_core.granularity.ipynb 5
+# %% ../../nbs/core/granularity.ipynb 5
 class Granularities:
     _granularities_Conv2d = {'weight':0, 'shared_weight':1, 'channel':2, 'column':3, 'row':4, 'kernel':(3,4), 'filter':(2,3,4), 'shared_channel':(1,2), 'shared_column': (1,3), 'shared_row': (1,4), 'vertical_slice': (2,3), 'horizontal_slice': (2,4), 'shared_vertical_slice': (1,2,3), 'shared_horizontal_slice': (1,2,4), 'shared_kernel': (1,3,4), 'layer':(1,2,3,4)}
     _granularities_ConvT2d = {'weight':0, 'shared_weight':2, 'channel':1, 'column':3, 'row':4, 'kernel':(3,4), 'filter':(1,3,4), 'shared_channel':(1,2), 'shared_column': (2,3), 'shared_row': (2,4), 'vertical_slice': (1,3), 'horizontal_slice': (1,4), 'shared_vertical_slice': (1,2,3), 'shared_horizontal_slice': (1,2,4), 'shared_kernel': (2,3,4), 'layer':(1,2,3,4)}

fasterai/core/schedule.py

@@ -1,17 +1,17 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/core/00c_core.schedules.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/core/schedules.ipynb.
 
 # %% auto 0
 __all__ = ['one_shot', 'iterative', 'agp', 'one_cycle', 'cos', 'lin', 'dsd', 'schedules', 'Schedule', 'sched_oneshot',
            'sched_iterative', 'sched_agp', 'sched_onecycle', 'sched_dsd', 'available_schedules']
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 2
+# %% ../../nbs/core/schedules.ipynb 2
 import numpy as np
 import matplotlib.pyplot as plt
 from fastcore.basics import *
 from fastai.callback.schedule import *
 import math
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 4
+# %% ../../nbs/core/schedules.ipynb 4
 class Schedule():
     "Base class to create schedules"
     def __init__(self, sched_func, start_pct=0., end_pct=None, start_sparsity=0.):
@@ -44,35 +44,35 @@ def plot(self, end_sparsity):
     def reset(self):
         self.current_sparsity, self.previous_sparsity = map(listify, [self.start_sparsity, self.start_sparsity])
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 9
+# %% ../../nbs/core/schedules.ipynb 9
 def sched_oneshot(start, end, pos): return end
 
 one_shot = Schedule(sched_oneshot, start_pct=0.5)
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 16
+# %% ../../nbs/core/schedules.ipynb 16
 def sched_iterative(start, end, pos, n_steps=3):
     "Perform iterative pruning, and pruning in `n_steps` steps"
     return start + ((end-start)/n_steps)*(np.ceil((pos)*n_steps))
 
 iterative = Schedule(sched_iterative, start_pct=0.2)
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 27
+# %% ../../nbs/core/schedules.ipynb 27
 def sched_agp(start, end, pos): return end + (start - end) * (1 - pos)**3
 
 agp = Schedule(sched_agp, start_pct=0.2)
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 32
+# %% ../../nbs/core/schedules.ipynb 32
 def sched_onecycle(start, end, pos, α=14, β=6):
     out = (1+np.exp(-α+β)) / (1 + (np.exp((-α*pos)+β)))
     return start + (end-start)*out
 
 one_cycle = Schedule(sched_onecycle)
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 38
+# %% ../../nbs/core/schedules.ipynb 38
 cos = Schedule(sched_cos)
 lin = Schedule(sched_lin)
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 42
+# %% ../../nbs/core/schedules.ipynb 42
 def sched_dsd(start, end, pos):
     if pos<0.5:
         return start + (1 + math.cos(math.pi*(1-pos*2))) * (end-start) / 2
@@ -81,7 +81,7 @@ def sched_dsd(start, end, pos):
 
 dsd = Schedule(sched_dsd)
 
-# %% ../../nbs/core/00c_core.schedules.ipynb 46
+# %% ../../nbs/core/schedules.ipynb 46
 schedules = ('one_shot', 'iterative', 'agp', 'one_cycle', 'cos', 'lin', 'dsd')
 def available_schedules():
     print(schedules)

fasterai/distill/distillation_callback.py

@@ -1,10 +1,10 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/distill/04_distill.knowledge_distillation.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/distill/distillation_callback.ipynb.
 
 # %% auto 0
 __all__ = ['KnowledgeDistillationCallback', 'get_model_layers', 'get_module_by_name', 'SoftTarget', 'Logits', 'Mutual',
            'Attention', 'ActivationBoundaries', 'FitNet', 'Similarity']
 
-# %% ../../nbs/distill/04_distill.knowledge_distillation.ipynb 18
+# %% ../../nbs/distill/distillation_callback.ipynb 18
 from fastai.vision.all import *
 
 import torch
@@ -14,7 +14,7 @@
 from functools import reduce
 from typing import Union
 
-# %% ../../nbs/distill/04_distill.knowledge_distillation.ipynb 20
+# %% ../../nbs/distill/distillation_callback.ipynb 20
 class KnowledgeDistillationCallback(Callback):
     def __init__(self, teacher, loss, activations_student=None, activations_teacher=None, weight=0.5):
         self.stored_activation_student, self.stored_activation_teacher  = {}, {}
@@ -59,7 +59,7 @@ def after_fit(self):
             self.remove_hooks(self.handles_t)
             self.remove_hooks(self.handles_st)
 
-# %% ../../nbs/distill/04_distill.knowledge_distillation.ipynb 22
+# %% ../../nbs/distill/distillation_callback.ipynb 22
 def get_model_layers(model, getLayerRepr=False):
     layers = OrderedDict() if getLayerRepr else []
     def get_layers(net, prefix=[]):
@@ -83,7 +83,7 @@ def get_module_by_name(module: Union[torch.Tensor, nn.Module],
     names = access_string.split(sep='.')
     return reduce(getattr, names, module)
 
-# %% ../../nbs/distill/04_distill.knowledge_distillation.ipynb 24
+# %% ../../nbs/distill/distillation_callback.ipynb 24
 def SoftTarget(pred, teacher_pred, T=5, **kwargs):
     return nn.KLDivLoss(reduction='batchmean')(F.log_softmax(pred/T, dim=1), F.softmax(teacher_pred/T, dim=1)) * (T*T)
 

fasterai/misc/bn_folding.py

@@ -1,15 +1,15 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/misc/06a_misc.bn_folding.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/misc/bn_folding.ipynb.
 
 # %% auto 0
 __all__ = ['BN_Folder']
 
-# %% ../../nbs/misc/06a_misc.bn_folding.ipynb 3
+# %% ../../nbs/misc/bn_folding.ipynb 3
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import copy
 
-# %% ../../nbs/misc/06a_misc.bn_folding.ipynb 12
+# %% ../../nbs/misc/bn_folding.ipynb 12
 class BN_Folder():
     def __init__(self):
         super().__init__()

fasterai/misc/cpu_optimizer.py

@@ -1,14 +1,14 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/misc/06c_misc.cpu_optimizer.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/misc/cpu_optimizer.ipynb.
 
 # %% auto 0
 __all__ = ['accelerate_model_for_cpu']
 
-# %% ../../nbs/misc/06c_misc.cpu_optimizer.ipynb 3
+# %% ../../nbs/misc/cpu_optimizer.ipynb 3
 import torch
 import torch.nn as nn
 from torch.utils.mobile_optimizer import optimize_for_mobile
 
-# %% ../../nbs/misc/06c_misc.cpu_optimizer.ipynb 4
+# %% ../../nbs/misc/cpu_optimizer.ipynb 4
 def accelerate_model_for_cpu(model: nn.Module, example_input: torch.Tensor):
     model.eval()
     example_input = example_input.to(memory_format=torch.channels_last)

fasterai/misc/fc_decomposer.py

@@ -1,15 +1,15 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/misc/06b_misc.fc_decomposer.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/misc/fc_decomposer.ipynb.
 
 # %% auto 0
 __all__ = ['FC_Decomposer']
 
-# %% ../../nbs/misc/06b_misc.fc_decomposer.ipynb 5
+# %% ../../nbs/misc/fc_decomposer.ipynb 5
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
 import copy
 
-# %% ../../nbs/misc/06b_misc.fc_decomposer.ipynb 6
+# %% ../../nbs/misc/fc_decomposer.ipynb 6
 class FC_Decomposer:
 
     def __init__(self):

fasterai/prune/prune_callback.py

@@ -1,9 +1,9 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/prune/03b_prune.prune_callback.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/prune/prune_callback.ipynb.
 
 # %% auto 0
 __all__ = ['PruneCallback']
 
-# %% ../../nbs/prune/03b_prune.prune_callback.ipynb 3
+# %% ../../nbs/prune/prune_callback.ipynb 3
 from fastai.vision.all import *
 from fastai.callback.all import *
 from .pruner import *
@@ -14,7 +14,7 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
-# %% ../../nbs/prune/03b_prune.prune_callback.ipynb 4
+# %% ../../nbs/prune/prune_callback.ipynb 4
 class PruneCallback(Callback):
     def __init__(self, sparsity:int, context:str, criteria:Callable, schedule:Callable, model:nn.Module=None, round_to:int=None, layer_type:nn.Module=nn.Conv2d):
         store_attr()

fasterai/prune/pruner.py

@@ -1,9 +1,9 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/prune/03_prune.pruner.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/prune/pruner.ipynb.
 
 # %% auto 0
 __all__ = ['Pruner']
 
-# %% ../../nbs/prune/03_prune.pruner.ipynb 2
+# %% ../../nbs/prune/pruner.ipynb 2
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -19,7 +19,7 @@
 from ..core.criteria import *
 from fastai.vision.all import *
 
-# %% ../../nbs/prune/03_prune.pruner.ipynb 4
+# %% ../../nbs/prune/pruner.ipynb 4
 class Pruner():
     def __init__(self, model, context, criteria, layer_type=[nn.Conv2d, nn.Linear, nn.LSTM], example_inputs=torch.randn(1,3,224,224), ignored_layers=None):
         store_attr()

fasterai/quantize/quantize_callback.py

@@ -1,9 +1,9 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/quantize/06b_quantize.quantize_callback.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/quantize/quantize_callback.ipynb.
 
 # %% auto 0
 __all__ = ['QuantizeCallback']
 
-# %% ../../nbs/quantize/06b_quantize.quantize_callback.ipynb 3
+# %% ../../nbs/quantize/quantize_callback.ipynb 3
 import copy
 import torch
 import torch.nn as nn
@@ -15,7 +15,7 @@
 import torch.ao.quantization.quantize_fx as quantize_fx
 from torch.ao.quantization.quantize_fx import convert_fx, prepare_fx
 
-# %% ../../nbs/quantize/06b_quantize.quantize_callback.ipynb 5
+# %% ../../nbs/quantize/quantize_callback.ipynb 5
 class QuantizeCallback(Callback):
     def __init__(self, qconfig_mapping=None, backend='x86'):
         self.qconfig_mapping = qconfig_mapping or get_default_qat_qconfig_mapping(backend)

fasterai/quantize/quantizer.py

@@ -1,9 +1,9 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/quantize/06_quantize.quantizer.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/quantize/quantizer.ipynb.
 
 # %% auto 0
 __all__ = ['Quantizer']
 
-# %% ../../nbs/quantize/06_quantize.quantizer.ipynb 2
+# %% ../../nbs/quantize/quantizer.ipynb 2
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -12,7 +12,7 @@
 import torch.ao.quantization.quantize_fx as quantize_fx
 from torch.ao.quantization.quantize_fx import convert_fx, prepare_fx
 
-# %% ../../nbs/quantize/06_quantize.quantizer.ipynb 4
+# %% ../../nbs/quantize/quantizer.ipynb 4
 class Quantizer():
     def __init__(self, backend="x86"):
         self.qconfig = get_default_qconfig_mapping(backend)

fasterai/regularize/regularize_callback.py

@@ -1,9 +1,9 @@
-# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/regularize/05_regularize.regularizer.ipynb.
+# AUTOGENERATED! DO NOT EDIT! File to edit: ../../nbs/regularize/regularize_callback.ipynb.
 
 # %% auto 0
 __all__ = ['RegularizeCallback']
 
-# %% ../../nbs/regularize/05_regularize.regularizer.ipynb 3
+# %% ../../nbs/regularize/regularize_callback.ipynb 3
 from fastai.callback.all import *
 from fastcore.basics import store_attr
 from ..core.criteria import *
@@ -13,7 +13,7 @@
 import torch.nn as nn
 import torch.nn.functional as F
 
-# %% ../../nbs/regularize/05_regularize.regularizer.ipynb 4
+# %% ../../nbs/regularize/regularize_callback.ipynb 4
 class RegularizeCallback(Callback):
     "Callback to apply grouped weight decay"
     def __init__(self, g, wd=0.01, layer_type=nn.Conv2d):