# Distilled model

A lightweight vision-language model based on the SigLIP architecture, extracting knowledge from the larger SigLIP-400m model through distillation techniques, suitable for zero-shot image classification tasks.

BERTA is obtained by distilling the embedding vectors of the FRIDA model into LaBSE-ru-turbo, which is used to calculate the embedding vectors of Russian and English sentences and supports multiple prefix tasks.

Flux.1 Lite is a distilled 8-billion parameter Transformer model derived from FLUX.1-dev, optimized for text-to-image generation tasks, reducing memory usage while maintaining accuracy and improving speed.

Kotoba-Whisper is a Japanese automatic speech recognition distilled Whisper model collection jointly developed by Asahi Ushio and Kotoba Technologies, which is 6.3 times faster than the original large-v3 while maintaining similar low error rates.

Distilled from the fine-tuned version of Stable-Diffusion v1-5's Dreamshaper v7, capable of generating high-quality images in extremely short inference time

IndicTrans2 is a high-quality machine translation model supporting translation between English and 22 Indian languages. This version is a distilled 200M parameter model

This model is a fine-tuned Arabic (Dz) to English translation model based on the distilled version of NLLB-200

SMaLL-100 is a compact and fast large-scale multilingual machine translation model, covering over 10,000 language pairs, with performance comparable to M2M-100 but smaller in size and faster.

This is a customized compact version of distilbert-base-multilingual-cased, specifically optimized for Polish while retaining the original model's accuracy.

This is a lightweight version of distilbert-base-multilingual-cased, specifically optimized for English and Italian, retaining the original model's accuracy.

This is a lightweight version of distilbert-base-multilingual-cased, supporting English, French, Spanish, Portuguese, and Italian language processing.