Understanding real-time model adaptability
Orca is powered by contextual embeddings in an external database, ensuring that your model uses the most appropriate data during inference.
Step 1: Teach the model to use context
Traditional neural networks learn to make predictions by memorizing the distribution of training data. However, this training data can fall out of date, compromising the model's effectiveness.
When you set-up a classifier or any other deep-learning model using Orca, the model learns to adapt outputs based on context you choose to supply.
Step 2: Access data during inference
Because the behavior is now encoded in the neural network, an Orca-trained model leverages that context during inference, just like a generative model using retrieval augmentation.
Now, any inference run reacts almost instantly to new information you supply to the model.
Step 3: Monitor. Update. Swap
Once your model learns to use Orca, you can proactively manage memories to address data drift and other dynamic changes. You can even do complete swaps of context for customized or personalized variants of a model.
Find out if Orca is right for you
Speak to our ML engineers to see if we can help you create more consistent models.