nam194/cribl-decision-1.0-merged

VISIONPricing:Input $0.4 / Cached $0.08 / Output $0.8Concurrent Unit Cost:1Model Size:4.5BQuant:BF16Context Size:32kTool Calling:SupportedPublished:Oct 15, 2026License:apache-2.0Architecture:Transformer Open Weights Featherless Exclusive Cold

nam194/cribl-decision-1.0-merged is a 4.5 billion parameter language model, a community-merged version of the cribl-ai/cribl-decision-1.0 LoRA adapter integrated into its Qwen/Qwen3.5-4B-Base foundation. This model is specifically designed for structured decision-making tasks, such as 'choice', 'noul', and 'score' types, by directly outputting probabilities for candidate options rather than generating free-form text. Its primary use case is efficient, structured classification and decision support, offering a compact solution without requiring PEFT dependencies.

Loading preview...

Overview

nam194/cribl-decision-1.0-merged is a 4.5 billion parameter model, representing a community-merged version of the cribl-ai/cribl-decision-1.0 LoRA adapter integrated into its base model, Qwen/Qwen3.5-4B-Base. This merge was performed using peft.PeftModel.merge_and_unload(), making it a standalone model that does not require PEFT dependencies to run. It is important to note that this is a community release and is not officially affiliated with Cribl AI or Qwen.

Key Capabilities

  • Structured Decision-Making: Unlike traditional text generation models, this model is engineered for structured readout. It does not generate free-form text but instead provides probabilities for predefined candidate options.
  • Decision Types: Supports specific decision types including choice, noul, and score.
  • Efficient Inference: By directly outputting candidate probabilities from the lm_head without full text generation, it offers a streamlined approach for classification tasks.
  • No PEFT Dependency: The merged architecture allows for direct loading and use without needing the PEFT library.

Good For

  • Classification Tasks: Ideal for scenarios requiring a model to select from a set of predefined options based on a given context.
  • Decision Support Systems: Can be integrated into applications that need to make structured decisions or score options.
  • Resource-Constrained Environments: As a 4.5B parameter model, it offers a relatively compact solution for structured inference compared to larger generative models.
  • Developers seeking a standalone solution: The merged weights eliminate the need for managing separate base models and adapters.

Finetunes
1 models