README: Kakeya Inference Engine for Mac — MLX spec-decode port journey (K3 beta baseline)#116
Closed
FluffyAIcode wants to merge 1 commit into
Closed
README: Kakeya Inference Engine for Mac — MLX spec-decode port journey (K3 beta baseline)#116FluffyAIcode wants to merge 1 commit into
FluffyAIcode wants to merge 1 commit into
Conversation
…y (K3 beta baseline) Records the decode-throughput journey from ~0.09x AR (O(T^2) collapse) -> ~0.2x (cross-runtime bridge) -> ~0.5x (all-MLX + CUDA-parity trim rollback) -> ~0.7x (block-4) -> ~1.0x (block-8, AR parity) on Gemma-4-26B-A4B / Mac M4, with each binding problem + fix, the ruled-out non-levers (quant/length/alignment/sync/ forced-over-gen artifact), the honest >AR-is-CUDA-favoured ceiling, and the evaluation environment (Mac bridge git-bus + self-hosted runner + evidence gate + H200). Recall 1.0 throughout; bounded S5 KV. Cross-links ADR 0009/0012/0013. Co-authored-by: FluffyAIcode <FluffyAIcode@users.noreply.github.com>
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
What
Adds a baseline record to
README.mdof the MLX-port throughput journey (your task 2), placed after the v0.3 GA evidence section.Contents
generate_step)RotatingKVCacherollback → CUDA-DynamicCacheparity (all-KVCache+ native trim, keep accepted)verify(L)compute — not rollback, sync count (single-fused probe stable; the 143 s pathology is large-cache-specific), acceptance, quantization, context length, or a missing alignment asset; the "2.13 low acceptance" was a forced-over-gen artifact. Remaining lever: the 4.5→7.7 drafter accept-len gap.kakeya-mac-m4runner), the evidence gate, and the H200 GPU side.Note on task 1 (beta merge + tag)
I could not merge PR #115 / create the
kakeya inference engine for mactag from the agent:ghis read-only,ManagePullRequesthas no merge action, and I won't push directly tomain. Also, PR #115 targets the b876 branch and stacks on the b876 all-MLX drafter (PR #112), neither onmain— so the Mac beta → main is a multi-PR consolidation (b876 #112 + #109/#110/#115) that the merge owner must perform, then tag the merge commit. Happy to prepare a consolidated beta-integration branch + PR-to-main if you want.Testing
Documentation-only — no code paths affected. (The numbers cited are from the real-Mac bridge runs recorded in PRs #109/#110/#115.)