Judge
Veritas - Reliability Judge
You can use the Veritas model to evaluate the factual correctness of model outputs, ensuring that the responses are accurate and free from hallucinated content.
We offer two versions of the Veritas model: Veritas and Veritas Nano. Veritas is a larger model that provides highly accurate and reliable judgments, making it ideal for batch processing applications. Veritas Nano is a streamlined version designed for real-time reliability checking with low latency.
Try it out for yourself here
Veritas
Introduction
Veritas is our state-of-the-art reliability judge designed for batch processing applications. As a larger model, Veritas provides highly accurate and super reliable judgments, making it ideal for scenarios where precision is paramount. In addition to providing labels, Veritas generates detailed rationales for its judgments, offering deeper insights into the decision-making process.Features
- High Accuracy: Achieves superior performance, ensuring reliable outcomes for complex tasks.
- Batch Processing Optimized: Best suited for processing large volumes of data where speed is less critical than thoroughness.
- Rationale Generation: Outputs detailed explanations alongside labels to clarify the reasoning behind each decision.
- Multi-turn Support: Effectively handles multi-turn interactions
Use Cases
- Content Analysis: Ideal for in-depth analysis of textual data in research or data mining projects.
- Quality Assurance: Employ in scenarios where understanding the rationale is crucial for validation and improvement processes.
Veritas Nano
Introduction
Veritas Nano is the streamlined version of our Veritas model, crafted for real-time reliability checking with a focus on low latency. Designed as a smaller model, it delivers quick and efficient judgments, making it perfect for applications that require immediate responses. Veritas Nano provides labels without additional rationale, streamlining the output for faster processing.Features
- Low Latency: Offers rapid response times with an average latency of just ~100ms.
- Real-time Processing: Optimized for environments where swift decision-making is critical.
- Efficient Performance: Maintains high reliability with an 85% accuracy rate despite its compact size.
- Multi-turn Compatibility: Supports multi-turn interactions
Use Cases
- Live Monitoring: Suitable for applications like live chat moderation where decisions need to be made instantaneously.
- User Interaction Platforms: Ideal for real-time applications such as customer service bots or interactive gaming.
Try it out for yourself here