Klyra-Decision-16M-MultiTask-Generalist-V5

Experimental Research Preview

Klyra-Decision-16M-MultiTask-Generalist-V5 is an experimental lightweight AI decision model developed by Polyvor Labs as part of the Klyra research initiative.

This release explores improvements in structured decision-making, multilingual understanding, and generalization across different tasks.

Current Status

Version 5 continues the development of the Klyra-Decision model family.

The model remains experimental and is intended for research, evaluation, and community feedback.

No particular benchmark accuracy is guaranteed.

Limitations

• May produce incorrect or inconsistent decisions.
• May struggle with ambiguous or unfamiliar tasks.
• Performance may decrease as the number of choices grows.
• Long-context processing may omit relevant evidence.
• Confidence scores may be unreliable outside evaluated data.
• Not recommended for production or high-stakes decisions.

Inference

Use klyra_v5_inference.py together with klyra_model.py for V5 inference.

The legacy inference helper does not enable V5 evidence-window processing.

Disclaimer

This model is provided for experimental research purposes.

---

Developed by Polyvor Labs.

Previous release: Klyra-Decision-16M-Generalist-V4