Translation for All
Demo
LLM
Gemma
LTEngine uses Google Gemma 3
LLM Upgrade
Traditionally LibreTranslate is based on Agros models, which have superseded by large language models (LLMs).
For example, using the 4 Chinese words 矢志奋斗 the LLM quality is much higher.
translate.aunsw.88.io output using argos models:
model.aunsw.88.io output using large language models:
API
Translation API example.
curl -v -X POST http://127.0.0.1:5050/translate \
-H "Content-Type: application/json" \
-d '{
"q": "Hello, world",
"source": "en",
"target": "es"
}'
Gemma 3
The LTEngine version of libretranslate can use many different LLMs, we are using Google's Gemma 3.
4-bit model
gemma-3-4b-it-q4_0.gguf
12-bit model
gemma-3-27b-it-q4_0.gguf
12-bit QAT model
gemma-3-12b-it-qat-q4_0-gguf
Although the result for 矢志奋斗 is the same, the Gemma 3 QAT (Quantization-Aware Training) models are supposed to be better than normal quantized 4-bit models because QAT maintains higher accuracy by simulating quantization during training.
Translation Integrated Applications
hub.aunsw.88.io
The translate.aunsw.88.io translation engine is a region wide resource and is being used by hub.aunsw.88.io
You can specify the language a post is written in and a "Translate" link will then be shown and can be clicked on to get the English translation.
Below image shows that a non-English language has been detected with a "Translate" link displayed:
After clicking on the "Translate" link, the Korean text has been converted to English.
