Run every loop at once

A parallel loop transformer as an editable map: shifted states, separate normalization, one batched call.

Preparing the preview…

Made to be adapted.

Loop 2 for token n reads loop 1’s output for token n−1, which was cached when that token was decoded. Every loop input exists as soon as token n arrives, so the loops run as one batch through the shared weights.

Data flow for one steady-state decoding step with two loops. Conceptual; no model is run.

Export formats: SVG / PNG / HTML
Editable in Studio: Explanation, appearance, and model inputs
Access: Free with attribution. No account needed.
Reference: Read the source

Loading your workspace…

Run every loop at once

Made to be adapted.

Another angle on the idea.

Draft, verify, correct

From base model to reasoning model

Run every loop at once

Made to be adapted.

Another angle on the idea.

Draft, verify, correct

From base model to reasoning model