gemma-3-1b-it-int4-web

Gemma 3 1B Instruct · int4 · Web (.task)

Not an official Google product. Google does not endorse this repository or the Lastly app.

This repository redistributes a single unmodified MediaPipe / LiteRT web artifact:

gemma3-1b-it-int4-web.task (~700 MB)

It is a browser (WebGPU) build of google/gemma-3-1b-it, as published by litert-community/Gemma3-1B-IT. We did not fine-tune, quantize, convert, or otherwise change the file.

Lastly 앱이 기기(브라우저)에서 문장을 나눌 때 쓰는 파일입니다. 가중치 학습·추가 양자화는 하지 않았습니다.

---

License / Terms

Gemma is provided under and subject to the Gemma Terms of Use.

By downloading, using, or redistributing the files in this repository, you agree to those terms. A copy of the notice required by Section 3.1 is in NOTICE. The full agreement is at the link above (and in LICENSE, if present).

Enforceable use restrictions (Gemma Terms §3.2)

You must not use Gemma or this artifact:

1. for the restricted uses in the Gemma Prohibited Use Policy; or
2. in violation of applicable laws and regulations.

If you distribute this file (or a derivative) further, you must also:

• pass these §3.2 restrictions on as an enforceable term;
• give recipients a copy of the Gemma Terms of Use;
• mark any files you modified;
• include a NOTICE file with this sentence:
  Gemma is provided under and subject to the Gemma Terms of Use found at ai.google.dev/gemma/terms

Outputs you generate are yours; Google claims no rights in them. You are responsible for how they are used.

Nothing here grants rights to Google trademarks, or implies Google endorsement.

---

Files

Provenance

---

Intended use

• In Lastly: on-device parse of short Korean household-log utterances (item name, date, interval) with MediaPipe LLM Inference in the browser. User text stays on the device; this file is not sent a user utterance by the app.
• Generally: MediaPipe / LiteRT web inference with WebGPU (Chrome, or Safari 26+).

Not intended as a general chatbot, a hosted API, or a substitute for professional advice.

Limitations

• 1B instruct model; answers can be wrong. Lastly treats output as a suggestion and asks the user to confirm before save.
• WebGPU only. CPU / Android .task variants are not in this repo.
• Quantized int4 web build; quality and speed differ from the full BF16 checkpoint.
• See the Gemma 3 model card for training data, evaluation, and safety notes.

---

How to load (web)

Direct file URL:

https://huggingface.co/nuuco/gemma-3-1b-it-int4-web/resolve/main/gemma3-1b-it-int4-web.task