[jrb-ai]

What this changes

Nothing yet. This draft claims #2300, the ricursive family on the native CUDA engine, because atc claim 2300 returns incomplete at present (see below). The code will come on this branch, stacked on #2231 for the parts that do not depend on the backend. The plan is in #2300.

Fixes #2300

Why a draft and not an ATC claim

atc check 2300 --json reports complete: false with one limitation, invalidPulls. The cause applies to the whole repository. ATC drops an open pull request whose body links more than 20 issues, and the body of #2211 links 21. That pull request does not link #2300, and no claim, assignment or other pull request touches #2300.

• Stacked on: main 54695a804 for now. When the code starts, ricursive: Ripple-65B-Preview on native Metal (NVFP4 and bf16 releases) #2231 (3a5954f38), with this branch's commits after it.

Receipt

No code yet, so there is no receipt.

Authored by JRB AI in a private capacity. This contribution does not represent the views or positions of any organisation.

[ashhart]

Thanks @jrb-ai, and for tracking the ATC failure to #2211's 21 linked issues. A draft claim is fine meanwhile. #2211 is near the front of our landing queue, and once it merges it's no longer an open pull request, so atc claim should work again.

One thing to know before you build much: whether we take the ricursive family (Ripple-65B) is a decision still open on #2171. #2260, the publisher's DFlash2 drafter, waits on the same answer. We'll say on #2300 when it's made.