feat: GF-T 2-layer backprop with scale_q eta (power-of-2) - #1861
Merged
Conversation
gft_xorbp2.t27: the full 2-layer XOR backprop step with scale_q applied to the weight updates -- eta=0.125=2^-3 folded into an exponent-offset shift, removing the 6 eta-multipliers. In-spec PASS; GF-T sim converges XOR 4/4 with eta=0.125. Size: single-core-with-sel backprop 23.7M -> 21.7M with scale_q, still over the ~17M ceiling (all 3 sel-branches synthesize into hardware). Finding: a stored-intermediates FSM is needed to fit -- forward once (store h,e,z,dz,x), then per-weight update frames reuse one shared update datapath. The sequential/coordinate backprop dynamic converges (verified in sim). Refs #1764 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Contributor
Contributor
|
📓 NotebookLM Notebook linked to this PR
This notebook contains session context, decisions, and artifacts for this work. |
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
gft_xorbp2.t27: full 2-layer XOR backprop with scale_q on the updates (eta=0.125=2^-3 -> exponent shift, removes 6 eta-multipliers). In-spec PASS; GF-T sim converges XOR 4/4. Size: single-core-with-sel backprop 23.7M->21.7M, still over the ~17M ceiling (all sel-branches in HW). Finding: a stored-intermediates FSM is needed to fit (forward once, shared update datapath per frame). Sequential/coordinate backprop dynamic converges. docs/NOW.md updated. Refs #1764