jylin001206/SII-Decision-4B-LoRA

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 1, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

jylin001206/SII-Decision-4B-LoRA is a 4.5 billion parameter decision model, adapted from Qwen/Qwen3.5-4B using LoRA, with a context length of 32768 tokens. It is specifically fine-tuned to return a probability distribution over user-supplied options, optimizing decision distributions rather than general chat. This model excels in tasks requiring probabilistic choices, binary/probabilistic Noul, and ordered Score tasks, making it suitable for applications focused on structured decision-making.

Loading preview...

SII-Decision-4B-LoRA Overview

SII-Decision-4B-LoRA is a specialized 4.5 billion parameter language model, fine-tuned from the Qwen/Qwen3.5-4B base model. Unlike general-purpose LLMs, this model is engineered to provide a probability distribution over user-defined options, focusing exclusively on decision-making tasks. It utilizes a LoRA adaptation (rank 64 / alpha 128) and has its weights merged for direct inference without requiring a separate adapter.

Key Capabilities

  • Optimized Decision-Making: Specifically trained to output probabilistic decisions across various structured tasks.
  • Structured Task Support: Handles Choice, binary/probabilistic Noul, and ordered Score tasks using a consistent answer-slot interface.
  • Efficient Inference: The complete merged weights can be loaded directly using transformers without PEFT, supporting BF16 precision.
  • Vision-Compatible Architecture: Retains the original Qwen3.5-4B vision components, though current decision training and inference primarily use text inputs.

Good For

  • Automated Decision Systems: Ideal for applications requiring a model to make probabilistic choices from a given set of options.
  • Structured Prediction Tasks: Excels in scenarios where the output is a score, a binary choice, or a probabilistic selection.
  • Research in Decision Modeling: Provides a specialized base for exploring and developing advanced decision-making AI.
  • Resource-Efficient Deployment: At 4.5B parameters, it offers a balance of capability and computational efficiency for specific decision-oriented workloads.

Finetunes
1 models