feat: GF-T full 2-layer XOR backprop step (learnable hidden + output) - #1860
Merged
Conversation
gft_xorbp.t27: full 2-layer backprop -- hidden W(2x2) AND output v(2) both trainable (fixed biases). Forward + MSE backprop through ReLU (dv_j=e*h_j, dz_j=e*v_j*relu'(z_j), dW_jk=dz_j*x_k). Returns the updated weight-PAIR by sel (u64: 0->hidden0, 1->hidden1, 2->output). A bit-faithful GF-T Python sim converges XOR 4/4 with BOTH layers learning (W->[[1,1],[1,1]], v->~[1,-2]) -- the capstone learnable-hidden-layer step (vs cycle 53's fixed-hidden). In-spec test PASS. Size: a naive 3-core wrapper is ~23.7K LUTs/core (redundant forward) -> over the ~17M ceiling; a shared-forward multi-frame structure is needed for silicon. Enabled by the cycle 56-57 ~2x magsub shrink. Refs #1764 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
Contributor
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
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
New spec gft_xorbp.t27 — full 2-layer backprop: hidden W(2x2) AND output v(2) both trainable. Forward + MSE backprop through ReLU. Returns updated weight-pair by sel (u64). A bit-faithful GF-T Python sim converges XOR 4/4 with BOTH layers learning (W->[[1,1],[1,1]], v->~[1,-2]) — the capstone learnable-hidden-layer step. In-spec PASS. Size: naive 3-core is ~23.7K LUTs/core (redundant forward), over the ~17M ceiling; shared-forward multi-frame needed for silicon. Enabled by cycle 56-57 2x magsub shrink. docs/NOW.md updated. Refs #1764