This model is a specialized fine-tune of Google's Gemma 3 4B, optimized specifically for the Polyglot Air desktop application workflow.
It has been trained to strictly adhere to suffix-based commands, ensuring instant, clean, and direct responses without conversational filler, hallucinations, or "Here is the translation" preambles.
🎯 Purpose
The default chat models often struggle with context switching when presented with short strings followed by a command (e.g., text::en). They might explain the translation or chat with the user.
This model is fine-tuned to treat the ::suffix as a strict instruction code.
- Input:
O arquivo não pode ser salvo porque o disco está cheio::en - Output:
The file cannot be saved because the disk is full
⚡ Suffix Command Table
Use these suffixes at the end of your selected text to trigger specific transformations.
| Suffix | Action | Example Input | Model Output |
|---|---|---|---|
::en | Translate to English | Bom dia::en | Good morning |
::pt | Translate to Portuguese (Portugal) | Good morning::pt | Bom dia |
::ptbr | Translate to Portuguese (Brazil) | The bus is coming::ptbr | O ônibus está vindo |
::es | Translate to Spanish | Hello friend::es | Hola amigo |
::zh | Translate to Chinese | Hello::zh | 你好 |
::fix | Fix grammar & spelling | i goes to skool yesterday::fix | I went to school yesterday |
::formal | Rewrite in a professional tone | Hey, send me that file asap::formal | Could you please send me that file as soon as possible? |
::casual | Rewrite in a casual/friendly tone | I acknowledge receipt of your message::casual | Got your message, thanks! |
::summary | Summarize the text | [Long Text]::summary | [Concise Summary] |
🚀 How to use with Polyglot Air
- Download this model (or the GGUF version if available).
- Add it to your Ollama library.
- Open Polyglot Air.
- Go to Settings > Model and select this model.
- Enjoy seamless, instruction-following translations.
Uploaded finetuned model
- Developed by: CalmState
- License: apache-2.0
- Finetuned from model : unsloth/gemma-3-4b-it-unsloth-bnb-4bit
This gemma3 model was trained 2x faster with Unsloth and Huggingface's TRL library.