jylin001206/SII-Decision-27B-Full

VISIONPricing:Input $1.6 / Cached $0.15 / Output $12Concurrent Unit Cost:2Model Size:27BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Oct 1, 2026License:apache-2.0Architecture:Transformer0.0K Open Weights Featherless Exclusive Cold

jylin001206/SII-Decision-27B-Full is a 27 billion parameter decision model, adapted from Qwen/Qwen3.8-27B, designed to return probability distributions over user-supplied options. It is specifically optimized for decision-making tasks, including choice, binary/probabilistic Noul, and ordered Score tasks, by training on 39,279 questions with full-rank updates to language linear layers. This model achieves a score of 80.1734/100 on a fixed 58-benchmark decision suite, making it suitable for applications requiring precise probabilistic decision outputs rather than general chat or SFT supervision.

Loading preview...

Overview

jylin001206/SII-Decision-27B-Full is a specialized 27 billion parameter model from the SII-Decision-27B family, built upon the Qwen/Qwen3.8-27B architecture. Unlike general-purpose language models, this model is exclusively designed to output a probability distribution over user-defined options, with its 'thinking' mechanism disabled. It is optimized for various decision tasks, including choice, binary/probabilistic Noul, and ordered Score tasks, all sharing a unified answer-slot interface.

Key Capabilities

  • Probabilistic Decision Outputs: Provides a probability distribution for user-supplied options, ideal for structured decision-making.
  • Specialized Training: Fine-tuned with full-rank updates on 39,279 questions, focusing on optimizing decision distributions rather than general chat or instruction following.
  • High Performance: Achieves a score of 80.1734 / 100 on a fixed 58-benchmark decision suite, indicating strong performance in its specialized domain.
  • Direct Inference: Can be loaded directly using the Hugging Face Transformers library without requiring PEFT or a separate base model, supporting BF16 precision.

Good For

  • Automated Decision Systems: Ideal for applications requiring a model to make probabilistic choices from a set of predefined options.
  • Structured Task Automation: Suitable for tasks like binary classification, Noul (no-opinion/uncertainty) tasks, and scoring scenarios where an ordered output is needed.
  • Research in Decision Modeling: Provides a robust base for exploring and developing advanced decision-making AI, with publicly available training data and code.

This model retains the Apache-2.0 license of the original Qwen weights.

Finetunes
1 models