Skip to content

Commit 364ea39

Browse files
author
semantic-release
committed
chore(release): 0.21.1
1 parent 69099e8 commit 364ea39

File tree

3 files changed

+39
-2
lines changed

3 files changed

+39
-2
lines changed

CHANGELOG.md

Lines changed: 37 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -1,6 +1,43 @@
11
# CHANGELOG
22

33

4+
## v0.21.1 (2026-03-09)
5+
6+
### Bug Fixes
7+
8+
- **postprocess**: Fix mypy type errors in _build_module_param_info
9+
([`11ea006`](https://github.com/johnmarktaylor91/torchlens/commit/11ea006c9a683675c926a9b0d649a2fa24b46558))
10+
11+
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
12+
13+
### Chores
14+
15+
- Trigger CI
16+
([`99f4102`](https://github.com/johnmarktaylor91/torchlens/commit/99f4102fa34564868291214d6b6cf3dfc239fdce))
17+
18+
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
19+
20+
### Performance Improvements
21+
22+
- **postprocess**: Optimize pipeline for large models
23+
([`a211417`](https://github.com/johnmarktaylor91/torchlens/commit/a2114170b4f5b441d4c0ab7add26435820fc8f8f))
24+
25+
- Per-step verbose timing: unwrap grouped _vtimed blocks into individual step timing with
26+
graph-stats summary, enabling users to identify which specific step is slow (O16) - Cache
27+
module_str by containing_modules tuple to avoid redundant string joins in Step 6 (O8) -
28+
Early-continue guards in _undecorate_all_saved_tensors to skip BFS on layers with empty
29+
captured_args/kwargs (O5) - Pre-compute buffer_layers_by_module dict in _build_module_logs,
30+
eliminating O(modules × buffers) scan per module (O6) - Single-pass arglist rebuild in Step 11
31+
rename, replacing 3-pass enumerate + index set + filter pattern (O2) - Replace OrderedDict with
32+
dict in _trim_and_reorder (Python 3.7+ preserves insertion order) for lower allocation overhead
33+
(O4) - Reverse-index approach in _refine_iso_groups: O(members × neighbors) instead of O(members²)
34+
all-pairs combinations (O9) - Pre-compute param types per subgraph as frozenset before pair loop
35+
in _merge_iso_groups_to_layers (O10) - Set-based O(n) collision detection replacing O(n²) .count()
36+
calls in _find_isomorphic_matches (O12)
37+
38+
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
39+
40+
441
## v0.21.0 (2026-03-09)
542

643
### Bug Fixes

pyproject.toml

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4,7 +4,7 @@ build-backend = "setuptools.build_meta"
44

55
[project]
66
name = "torchlens"
7-
version = "0.21.0"
7+
version = "0.21.1"
88
description = "A package for extracting activations from PyTorch models"
99
readme = "README.md"
1010
license = "GPL-3.0-only"

torchlens/__init__.py

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -9,7 +9,7 @@
99
first imported. See the bottom of this file for details.
1010
"""
1111

12-
__version__ = "0.21.0"
12+
__version__ = "0.21.1"
1313

1414
# ---- Public API: user-facing entry points --------------------------------
1515

0 commit comments

Comments
 (0)