gemma-4-e2b-web-mtp-vision-litert-lm

jfan/gemma-4-e2b-web-mtp-vision-litert-lm

gemma-4-e2b-web-mtp-vision-litert-lm is a text generation model from jfan. Use it when you need the model to write or continue text. It is set up for litert. The card lists the license as gemma.

Web-optimized gemma-4-E2B-it LiteRT-LM bundles for browser (WebGPU) execution, built with the official conversion pipeline (exportvit, convertermain, litertlmbuildercli).

Downloads · 30 days

Access

Public

Updated Oct 29, 2026

Repo size

83.9 GB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.litertlm8.9 GB · 98%

At a glance

Task: Text Generation
Library: litert
License: gemma
Access: Public
Created: Sep 12, 2026
Updated: Oct 29, 2026
SHA: 57e71c00

Try a prompt

Model card

From the Hugging Face model README

Gemma 4 E2B — WebGPU Optimized with Neural Multi-Token Prediction (MTP), Vision (1120 tokens), and Audio

Web-optimized gemma-4-E2B-it LiteRT-LM bundles for browser (WebGPU) execution, built with the official conversion pipeline (export_vit, converter_main, litertlm_builder_cli).

📦 Canonical Model Variants

Variant File	Size (Bytes)	SHA-256
`gemma-4-E2B-it-web-mtp.litertlm`	2,054,365,440	`394ec560d01909286bc9c3d2fbea1e7ac9744706a8f259107ec5644e966b3b76`
`gemma-4-E2B-it-web-mtp-vision.litertlm`	2,288,982,644	`e7f2078ec0e3c05542b7a89dff90c9442ed524aedc88097d53b44d8bac73550c`
`gemma-4-E2B-it-web-mtp-audio.litertlm`	2,157,894,260	`d118e3480b190eefa9ea6d8d3713e0ad864a258f8cfa8253bc42925dc52bfe57`
`gemma-4-E2B-it-web-mtp-vision-audio.litertlm`	2,392,513,140	`ff906922df5aafcd7322cd02bb6ceb478a64e4426c9b2991d5f0586dccfd75a1`

👁️ Multimodal Visual Token Budgets

Multi-signature vision_encoder (vision_70, vision_140, vision_280, vision_560, vision_1120) exported via //third_party/odml/generative/modules/gemini/export:export_vit with QAT Int8 weight recovery and query-chunked attention (max_query_chunk_size = 630, $k=16$ at vision_1120).
Artisan decoder vision_tokens_num: 1121 (tokens_num = 1120).

🎙️ Multimodal Audio Support

Audio Encoder: audio_encoder_hw (Conformer USM, FP16 preferred)
Audio Adapter: audio_adapter (CPU constraint)
End of Audio: end_of_audio soft-token embedding

Task: Text Generation
Library: litert
License: gemma
Created: Sep 12, 2026
Updated: Oct 29, 2026