By the numbers

A factual snapshot of the repository as of commit c923d169b1 (April 30, 2026), main branch.

Size

Metric	Value
Total commits in history	103,407
First commit	2012-01-25
Latest commit	2026-04-30
Repo age	~14 years
Top-level directories	~70
Tracked files	~75K

Lines of code (excluding build/, third_party/, .git/)

Language	LOC
Python	569,640
C++ (.cpp)	860,369
C/C++ headers	398,685
CUDA (.cu)	86,531
Total (these)	~1.92M

(Plus YAML, CMake, BUCK/Bazel, shell, Markdown — ~100K extra.)

Activity (last 6 months)

Metric	Value
Commits	~8,183
Unique committer emails	860
Approximate weekly throughput	~315 commits

Bot share

Metric	Value
Total commits with bot/noreply in author email	10,564
Bot share of total history	~10.2%

The largest single source of bot commits is the merge bot (facebook-github-bot, then pytorchbot) landing PRs through ghstack. Most of the human-authored work shows up in the merge commit.

Top-level directory sizes

Directory	Approx LOC	Notes
aten/	~600K	ATen tensor library (C++)
torch/	~1.3M	Python + torch/csrc/ C++
c10/	~70K	Core C++
tools/	~80K	Codegen, build, lint
torchgen/	~20K	The codegen Python package
test/	~700K	Test suite
benchmarks/	~50K	Microbenchmarks + nightly perf jobs
third_party/ (gitmodules)	~50 modules	NCCL, FBGEMM, kineto, cutlass, …

Largest files

File	Approx LOC	Notes
torch/_torch_docs.py	~12,000	Docstrings for every torch.* op
torch/__init__.py	~3,500	Public surface
aten/src/ATen/native/native_functions.yaml	~17,000	The op registry
tools/autograd/derivatives.yaml	~3,000	Backward formulas
.lintrunner.toml	~3,000	Lint configuration

Test surface

Layer	Approx files
test/	1,000+ Python test files
test/inductor/	~100 inductor-specific tests
test/distributed/	~150 distributed tests
aten/src/ATen/test/	~80 C++ unit tests
c10/test/	~40 C++ unit tests
test/cpp/	~60 C++ tests for the C++ frontend

Where to read next

• Lore — major eras of the project.
• Fun facts — anecdotes & curiosities.
• Deployment — release & CI topology.