Skip to content

feat: add transformer head pruning to Pruner #29

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
nathanhubens wants to merge 3 commits into
base: master
Choose a base branch
from
Open

feat: add transformer head pruning to Pruner #29

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
f054afa
feat: add transformer head pruning to Pruner
nathanhubens Apr 9, 2026
71caed5
fix: auto-patch timm Attention.forward for head pruning compatibility
nathanhubens Apr 9, 2026
60adecd
fix: auto-patch timm Attention.forward, add head freeze, add tutorials
nathanhubens Apr 10, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
42 changes: 38 additions & 4 deletions fasterai/_modidx.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -208,6 +208,34 @@
'fasterai/export/onnx_exporter.py'),
'fasterai.export.onnx_exporter.verify_onnx': ( 'export/onnx_exporter.html#verify_onnx',
'fasterai/export/onnx_exporter.py')},
'fasterai.huggingface.huggingface': { 'fasterai.huggingface.huggingface.HFSparsifyCallback': ( 'huggingface/huggingface.html#hfsparsifycallback',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback.__init__': ( 'huggingface/huggingface.html#hfsparsifycallback.__init__',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback._sparsity_value': ( 'huggingface/huggingface.html#hfsparsifycallback._sparsity_value',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback.on_epoch_end': ( 'huggingface/huggingface.html#hfsparsifycallback.on_epoch_end',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback.on_log': ( 'huggingface/huggingface.html#hfsparsifycallback.on_log',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback.on_optimizer_step': ( 'huggingface/huggingface.html#hfsparsifycallback.on_optimizer_step',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback.on_step_begin': ( 'huggingface/huggingface.html#hfsparsifycallback.on_step_begin',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback.on_train_begin': ( 'huggingface/huggingface.html#hfsparsifycallback.on_train_begin',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.HFSparsifyCallback.on_train_end': ( 'huggingface/huggingface.html#hfsparsifycallback.on_train_end',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface._has_transformers': ( 'huggingface/huggingface.html#_has_transformers',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface._load_model': ( 'huggingface/huggingface.html#_load_model',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface._require_transformers': ( 'huggingface/huggingface.html#_require_transformers',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface._save_compressed': ( 'huggingface/huggingface.html#_save_compressed',
'fasterai/huggingface/huggingface.py'),
'fasterai.huggingface.huggingface.sparsify_model': ( 'huggingface/huggingface.html#sparsify_model',
'fasterai/huggingface/huggingface.py')},
'fasterai.misc.all': {},
'fasterai.misc.bn_folding': { 'fasterai.misc.bn_folding.BN_Folder': ( 'misc/bn_folding.html#bn_folder',
'fasterai/misc/bn_folding.py'),
Expand Down Expand Up @@ -259,12 +287,18 @@
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner._build_pruning_schedule': ( 'prune/pruner.html#pruner._build_pruning_schedule',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner._detect_attention_heads': ( 'prune/pruner.html#pruner._detect_attention_heads',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner._freeze_head_pruning': ( 'prune/pruner.html#pruner._freeze_head_pruning',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner._patch_attention_forward': ( 'prune/pruner.html#pruner._patch_attention_forward',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner._resolve_pruning_ratio_dict': ( 'prune/pruner.html#pruner._resolve_pruning_ratio_dict',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner._sync_attention_attrs': ( 'prune/pruner.html#pruner._sync_attention_attrs',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner._to_tp_scheduler': ( 'prune/pruner.html#pruner._to_tp_scheduler',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner.get_attention_layers_to_ignore': ( 'prune/pruner.html#pruner.get_attention_layers_to_ignore',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner.get_ignored_layers': ( 'prune/pruner.html#pruner.get_ignored_layers',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner.get_linear_layers_to_ignore': ( 'prune/pruner.html#pruner.get_linear_layers_to_ignore',
Expand All @@ -275,8 +309,8 @@
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner.prune_model': ( 'prune/pruner.html#pruner.prune_model',
'fasterai/prune/pruner.py'),
'fasterai.prune.pruner.Pruner.restore_attention_layers': ( 'prune/pruner.html#pruner.restore_attention_layers',
'fasterai/prune/pruner.py')},
'fasterai.prune.pruner._pruning_compatible_forward': ( 'prune/pruner.html#_pruning_compatible_forward',
'fasterai/prune/pruner.py')},
'fasterai.quantize.all': {},
'fasterai.quantize.quantize_callback': { 'fasterai.quantize.quantize_callback.QuantizeCallback': ( 'quantize/quantize_callback.html#quantizecallback',
'fasterai/quantize/quantize_callback.py'),
Expand Down
1 change: 1 addition & 0 deletions fasterai/prune/prune_callback.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -53,6 +53,7 @@ def before_fit(self) -> None:
criteria=self.criteria,
pruning_ratio=self.pruning_ratio,
context=self.context,
example_inputs=self.example_inputs,
iterative_steps=total_training_steps,
schedule=pruning_schedule,
*self.extra_args,
Expand Down
Loading
Loading