latentdivergence/domain-expansion-gemma-4-e2b-vision

Domain Expansion Gemma 4 E2B — Android, vision

A verified mirror of the multimodal LiteRT-LM Gemma 4 E2B artifact used by Domain Expansion XR for the room scan: the same model answers questions about a camera frame as well as text.

• —File: gemma-4-E2B-it.litertlm
• —Size: 2,588,147,712 bytes
• —SHA-256: 181938105e0eefd105961417e8da75903eacda102c4fce9ce90f50b97139a63c
• —Runtime: LiteRT-LM for Android; text decoder and vision encoder on the GPU backend
• —Intended hardware: Android arm64 devices with a compatible GPU
• —Inference: fully on-device after the one-time download; nothing is sent anywhere

Provenance

This artifact is mirrored byte-for-byte from litert-community/gemma-4-E2B-it-litert-lm (gemma-4-E2B-it.litertlm). It is provided under Apache-2.0, as published there. The text-only GPU export the app used before this lives at latentdivergence/domain-expansion-gemma-4-e2b-mobile.

Use

Download gemma-4-E2B-it.litertlm to app-private storage. Do not package it in an APK. Initialise it lazily with LiteRT-LM using Backend.GPU() for both the decoder and visionBackend, send one encoded JPEG per turn as Content.ImageBytes, and release the engine when the companion is idle.