chinhnc/Quyet-1.0-Medium
Quyet-1.0-Medium by Chinh Nguyen is a 4.66 billion parameter decision model based on Qwen3.5-4B, fine-tuned with a LoRA adapter. It specializes in taking a text-based state and multiple typed questions (choice, score, noul) to provide calibrated probabilistic answers. This model is designed for structured decision-making tasks, supporting English and Vietnamese, and can process states up to 6,000 tokens within an 8,000-token prompt.
Loading preview...
Quyet-1.0-Medium: A Calibrated Decision Model
Quyet-1.0-Medium, developed by Chinh Nguyen, is a specialized decision model built upon the Qwen3.5-4B architecture. It features 4.66 billion parameters (4.21B text) and is fine-tuned with a LoRA adapter to excel at structured decision-making. Unlike general-purpose LLMs, Quyet-1.0-Medium processes a given text state (e.g., conversation, JSON) and responds to specific, typed questions with calibrated probabilities.
Key Capabilities
- Structured Decision-Making: Provides precise answers to
choice,score, andnoul(true/false) question types. - Calibrated Probabilities: Returns confidence scores and probability distributions for each decision, enhancing reliability.
- Multilingual Support: Primarily tuned for English and Vietnamese, with functional support for other languages.
- Flexible Input: Handles states up to 6,000 tokens within an 8,000-token prompt, allowing for complex contexts.
- Efficient Inference: Runs in bf16 on a 16 GB GPU, utilizing a unique letter-readout decision prompt mechanism.
Good for
- Automated Customer Support: Classifying customer intent, urgency, or sentiment from messages.
- Data Extraction & Categorization: Extracting structured decisions from unstructured text.
- Business Process Automation: Guiding workflows based on specific criteria and confidence levels.
- Any application requiring probabilistic, structured answers from text inputs.
A live demo of the Quyet-1.0-Large model is available at quyet.ai.