Cam SmithAI Studio

← All models

Computational Model Studio · Model release

Gemma 3n E2B IT LiteRT preview

Produces text from text or images with a Google AI Edge preview checkpoint prepared for Android deployment.

Developer: Google DeepMind · Last verified:

What it is designed to do

Generate text responses to text prompts and image-and-text questions with the published Google AI Edge preview checkpoint.

Primary contribution
generation transformation
Input
text, image
Output
text (generated text, visual question answer)

Documented uses: Android text chat and image-question tasks in the Google AI Edge Gallery allowlist.

Access and deployment

  • Gemma 3n E2B IT LiteRT int4 task bundle — restricted. Hugging Face sign-in and acceptance of Google's Gemma usage terms are required to download the task bundle. Artifact documentation
    • mobile edge: Compatible Android device with CPU or GPU accelerator and sufficient free memory for the estimated peak; about 5.91 GB estimated peak RAM; Google AI Edge Gallery allowlist configuration: 4096 maximum tokens, text chat or image question; file size is about 3.14 GB and estimated peak memory is 5,905,580,032 bytes. (publisher stated estimate) Limits: The allowlist reports an estimated peak; the card's separate Samsung S25 Ultra benchmark measures different conditions and cannot replace this estimate; The preview checkpoint supports text and vision only; audio and video are not established for this artifact; Memory may vary by Android device, runtime and workload.
    Gemma Terms of Use.
Technical profile and review notes
Construction
learned
Representation
neural
Normalized architecture tags
No normalized architecture tag assigned
Lifecycle
experimental

Inclusion rationale: Adds a publisher-listed Android edge bundle with an exact artifact and configuration-specific peak-memory estimate.; Adds bounded on-device text-and-image-to-text use without borrowing broader family modalities.

Limitations

  • This LiteRT preview checkpoint supports text and vision input only; the wider Gemma 3n family's audio and video capabilities must not be projected onto it.
  • The Google AI Edge Gallery peak-memory figure is an estimate for its configuration, not a universal requirement; the card separately reports a measured Samsung S25 Ultra benchmark under different settings.
  • The publisher labels this repository a preview, so capabilities and deployment behavior require re-verification before public release.

Sources and related Studio records

Sources

  1. Google Gemma 3n E2B LiteRT preview model cardGoogle DeepMind · checked 2026-09-30
  2. Google AI Edge Gallery model allowlistGoogle AI Edge · checked 2026-09-30
  3. Google Gemma 3n E2B LiteRT preview filesGoogle DeepMind · checked 2026-09-30
  4. Google Gemma Terms of UseGoogle · checked 2026-09-30

Connections