chinhnc/Quyet-1.0-Large
Quyet-1.0-Large by Chinh Nguyen is a 31.3 billion parameter decision model based on the Gemma-4-31B-it architecture, fine-tuned with a merged LoRA. It specializes in taking a text, JSON, or conversation state and one or more typed questions to pick options with calibrated probabilities. This model excels at structured decision-making tasks, supporting choice, score, and true/false question types, and is tuned for English and Vietnamese with a 32,768 token context length.
Loading preview...
Quyet-1.0-Large: A Calibrated Decision Model
Quyet-1.0-Large, developed by Chinh Nguyen, is a 31.3 billion parameter model designed specifically for structured decision-making. Built upon the google/gemma-4-31B-it base model, it incorporates a merged LoRA fine-tune and a unique letter-readout decision prompt mechanism. This model processes a given state (text, JSON, or conversation) and multiple typed questions to return a selected option for each question, along with calibrated probabilities.
Key Capabilities
- Decision-Making: Predicts one option per question from a given state, providing calibrated probabilities for each choice.
- Question Types: Supports
choice(select one label),score(select from an ordered scale), andnoul(true/false) question formats. - Multilingual Support: Primarily tuned for English and Vietnamese, with functionality for other languages at potentially lower accuracy.
- Context Handling: Processes input states up to 6,000 tokens within an 8,000-token prompt, with conversation lists prioritizing recent turns.
- Output Format: Answers follow the TypeSafe
/v1/systemoneshape, including probabilities forchoiceandscoretypes, and P(true) fornoul.
Good For
- Automated customer service intent classification.
- Sentiment analysis with calibrated confidence scores.
- Structured data extraction requiring probabilistic choices.
- Any application needing a model to make explicit, explainable decisions from complex inputs.