Polyglot r2: Suffix-Based Command Model

CalmState/Qwen-3-4b-Polyglot-r2 is a specialized fine-tune of Qwen-3 4b, developed by CalmState, engineered for strict adherence to suffix-based commands. This 4 billion parameter model, with a 40960 token context length, is optimized for the Polyglot Air desktop application workflow. It was trained on a dataset approximately 4 times larger than its predecessor, significantly enhancing its command range and response accuracy. The model is designed to eliminate conversational filler and hallucinations, providing clean, direct outputs for specific instructions.

Key Capabilities

• Strict Instruction Following: Processes ::suffix as a direct command, avoiding preambles or explanations.
• Multilingual Translation: Supports translation to English (::en), Portuguese (::pt, ::ptbr), Spanish (::es), and Chinese (::zh).
• Tone & Style Adjustment: Rewrites text for formality (::formal, ::casual, ::informal, ::polite), technicality (::technical), creativity (::creative), business (::business), news (::news), and social media (::social).
• Content Manipulation: Performs tasks like summarizing (::summarize), expanding (::expand, ::elaborate), simplifying (::simplify), making concise (::concise), and transforming into questions (::toQuestion) or statements (::toStatement).
• Grammar & Spelling Correction: Includes a ::fix command for correcting errors.

Good for

• Developers integrating precise, command-driven text transformations into applications like Polyglot Air.
• Use cases requiring instant, non-conversational outputs for translation or text manipulation.
• Automated workflows where strict adherence to input commands is critical to avoid unwanted conversational elements.