Premonition-Exp architecture graft and assembly tooling
Copyright 2026 Prem Research.

Contains unchanged DiffusionGemma backbone and Gemma 4 E4B audio tower weights.
A lightweight audio projector was trained for the Gemma 4 audio graft.
Source model identifiers and immutable revisions appear in assembly_manifest.json.

Hugging Face Transformers: Copyright The Hugging Face team.
Gemma 4 / DiffusionGemma model weights: Google / Google DeepMind.
vLLM: Copyright contributors to the vLLM project.
Not affiliated with or endorsed by Google or TypeSafe AI.