Now / public snapshot
11 July 2026 · Tel Aviv

What is actually on the bench.

A hand-curated record of the current focus and public open loops. This is intentionally dated; it is a working state, not a claim of permanent priority.

Primary focus / 01

Make the MTP drafter smaller than the overhead it is meant to remove.

The primary study is StudentSV: a half-width MTP block with its own 32,768-token output head, trained for the real speculative-decoding path instead of inheriting the target model's full 248,320-row projection.

The current run retains 78% of the windowed teacher's code acceptance at roughly 13% of its estimated draft FLOPs. That is a reason to scale the experiment, not yet a speedup claim. The next gate is recursive acceptance and measured end-to-end wall time.

Subject
Qwen3.5-9B · co-trained MTP → StudentSV
Methods
Small-vocabulary head · half-width block · replay acceptance
Standard
Separate training signal, acceptance, FLOPs, and measured wall time.

Working on the same edge? Compare notes.