IMP-VISION β Curated Model Sets
Deadbug Labs' IMP-VISION smart glasses run owner-chosen models: any compatible Hugging Face repo can fill each role (Eyes / Brain / Ears / Voice) in the glasses' picker. This repo is the curated index β the hand-verified defaults and the RAM-checked sets the app offers first. Nothing here re-hosts third-party weights; every entry points at the upstream repo, with its license, exactly as the picker shows it before download.
Shipping defaults (Balanced set, ~1.7 GB peak)
- Eyes β ggml-org/SmolVLM-500M-Instruct-GGUF β 929 MB with mmproj, Apache-2.0
- Brain β Qwen/Qwen2.5-1.5B-Instruct-GGUF β 1,117 MB, Apache-2.0
- Ears β ggerganov/whisper.cpp
ggml-base.enβ 148 MB, MIT - Voice β rhasspy/piper-voices
en_US-lessac-mediumβ 63 MB ONNX
Sets
- Light (~1.4 GB peak) β SmolVLM-256M Β· Llama-3.2-1B Β· whisper base.en Β· Piper lessac
- Balanced (~1.7 GB peak) β the defaults above
- Max (~3.4 GB peak) β larger Eyes/Brain picks for high-RAM phones
The full format-checked catalog behind these picks indexes 1,242 compatible repos (219 Eyes / 867 Brain / 75 Ears / 81 Voice). The picker accepts any of them β the sets are recommendations, not a whitelist.
By Deadbug Labs β open hardware boards and the devices built on them.
Find Deadbug Labs: GitHub Β· Facebook Β· Hugging Face Β· Site