morriszjm/Tacit-8B
Tacit-8B, developed by morriszjm, is an 8 billion parameter model built on Qwen3-8B, designed for efficient typed decision-making in a single forward pass. It utilizes AnyJev self-distillation, where the base model generated and answered its own decision problems, learning to provide direct probabilistic outputs for options, yes/no, or ordered levels. This unique training approach allows Tacit-8B to achieve improved accuracy on benchmarks like JevBench and bev-decision compared to its base model, making it suitable for applications requiring fast, structured decision outputs.
Loading preview...
Tacit-8B Overview
Tacit-8B is an 8 billion parameter language model, based on Qwen/Qwen3-8B, specifically engineered for efficient typed decision-making in a single forward pass. This model can choose an option, answer yes/no, or select an ordered level, returning a probability distribution over the possible choices.
Key Capabilities
- Single-Pass Decision Making: Designed to make structured decisions (e.g., multiple choice, binary, ordinal) in one forward pass, providing a probability over options.
- Self-Distillation Training: Uniquely trained using AnyJev self-distillation, where the base Qwen3-8B model generated its own decision problems and learned to provide direct answers without human labels or external datasets.
- Improved Accuracy: Demonstrates enhanced accuracy on decision-making benchmarks compared to its Qwen3-8B base, achieving +3.9 points on JevBench (public set) and +1.4 points on bev-decision (test split).
Good for
- Applications requiring fast, structured decision outputs.
- Scenarios where probabilistic choices over predefined options are needed.
- Use cases benefiting from a model trained without reliance on external human-labeled datasets for decision tasks.