feat: Add minimal transformer block with multi-head attention - #10
Open
ThomasHartDev wants to merge 2 commits into
Open
feat: Add minimal transformer block with multi-head attention#10ThomasHartDev wants to merge 2 commits into
ThomasHartDev wants to merge 2 commits into
Conversation
Pre-LN encoder block: multi-head self-attention, position-wise GELU FFN, residual connections, and LayerNorm. Numpy only, with shape and mask tests.
Document transformer as forward-only (no optimizer/grad claim). Pin scaled_dot_product_attention to softmax(QK^T/sqrt(d_k))V with a scale-sensitive golden test. Softmax all-(-inf) rows return zeros without RuntimeWarning/nan; fully-masked attention covered in tests.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Add a forward-only Pre-LN self-attention block (multi-head scaled dot-product attention, position-wise GELU FFN, residual + LayerNorm) with optional causal masking.
No backward/grad path yet:
parameters()exposes weight tensors for inspection only, not for the MLP optimizers. Attention is checked against a goldensoftmax(QKᵀ/√d_k)Vreference (scale load-bearing). All-masked (-inf) softmax rows return zeros without RuntimeWarning/nan.