Skip to content

master#58

Merged
Eamon2009 merged 164 commits into
expfrom
master
Jul 23, 2026
Merged

master#58
Eamon2009 merged 164 commits into
expfrom
master

Conversation

@Eamon2009

Copy link
Copy Markdown
Member

No description provided.

dependabot Bot and others added 30 commits July 14, 2026 15:10
Bumps [postcss](https://github.com/postcss/postcss) from 8.5.15 to 8.5.19.
- [Release notes](https://github.com/postcss/postcss/releases)
- [Changelog](https://github.com/postcss/postcss/blob/main/CHANGELOG.md)
- [Commits](postcss/postcss@8.5.15...8.5.19)

---
updated-dependencies:
- dependency-name: postcss
  dependency-version: 8.5.19
  dependency-type: direct:development
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>
Eamon2009 added 25 commits July 21, 2026 17:56
…n.cu` to serve as the CUDA-accelerated driver for dataset loading, GPU initialization, backprop training, chat sessions, and token generation.
…ation and added new directories for CI/CD workflows and assets.

Updated file structure section in README.md to reflect changes in project organization and added new directories for CI/CD workflows and assets.
Introduce foundational backpropagation gradient structures and primitive
backward operations for transformer layer components.

Key additions:
- Gradient Accumulation Structures: Define `Grads` root container and child
  structs (`GradLinear`, `GradEmbedding`, `GradLayerNorm`, `GradHead`, `GradMHA`,
  `GradFFN`, `GradBlock`) equipped with zero-initialization helpers.
- Activation Memory Structs: Define `SavedForward` and layer-specific intermediate
  tensors needed for reverse-mode automatic differentiation.
- Primitive Backward Kernels:
  - `backward_cross_entropy`: Evaluates softmax derivatives scaled by batch size ($1 / BT$).
  - `backward_linear`: Computes input gradients ($dX = dOut \cdot W^T$) and
    accumulates weight/bias gradients ($dW$, $db$).
  - `backward_layernorm`: Implements full LayerNorm backprop incorporating mean/std
    jacobian terms along sequence channels.
  - `backward_relu`: Computes element-wise activation derivative masks.
…nhance comment formatting. Adjusted steps for extracting PR and comment metadata, and refined success/failure reporting.

Updated PR comment checks workflow
Updated project structure section with detailed descriptions for each directory and file.
@Eamon2009
Eamon2009 merged commit 5d0f109 into exp Jul 23, 2026
24 of 39 checks passed
@Eamon2009
Eamon2009 requested review from a team and eamonsippy and removed request for a team July 23, 2026 14:55
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