Skip to content

Basic Autograd engine. - #9

Merged
sourcepirate merged 4 commits into
mainfrom
feat/autograd
Jun 30, 2026
Merged

Basic Autograd engine.#9
sourcepirate merged 4 commits into
mainfrom
feat/autograd

Conversation

@sourcepirate

Copy link
Copy Markdown
Owner

No description provided.

…tensor_loss

- Clean up base_attention.py: remove dead _numpy_attention method and
  the one-line _tensor_attention wrapper, inlining the implementation
  directly into scaled_dot_product_attention()
- Remove dead _tensor_loss method from losses/base.py
- No functional changes — 379/379 tests pass
- Add docs/README.md — comprehensive documentation index mapping
  every component to its documentation and original research paper
  with arXiv links
- Update README.md — add prominent 'Documentation' section linking
  to docs/README.md
- Update 'Deep Dives' section with added Autograd Engine link
@sourcepirate

Copy link
Copy Markdown
Owner Author

/oc can you fix the tests ?

…description

The project now has a full autograd engine built from scratch in NumPy
(Tensor, GradientTape, ops). Most layers use tape-based backward via
base.Layer.backward() rather than manual backward methods.
@sourcepirate
sourcepirate merged commit 24fbb50 into main Jun 30, 2026
3 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant