IGroundedConversationTrainable<T>
Interfaces · AiDotNet.VisionLanguage.Interfaces
A model that trains on grounded conversations: an image, a conversation about it, the regions its questions name, and
a mask for every [SEG] in its answers.
Known implementations: GLaMM<T>
How It Works
Such a model’s loss combines the language-model loss with mask losses; a tensor target cannot carry the
conversation, so these models refuse Train(input, expected) and train here instead.
Methods
| Method | Summary |
|---|---|
T EvaluateConversationLoss(GlammSample<T> sample) |
The paper loss on the same sample, without training. |
void TrainConversation(GlammSample<T> sample) |
One training step with the model’s paper loss. |