Repository navigation
Train() needs to delineate between examples and feedback #14
Description
Something needs to be sent back the JSON for train that distinguishes between examples and feedback so that the agent can condition on this if necessary. For example an agent that has different memory mechanisms depending on these cases.
Activity
eharpste commentedon Dec 7, 2020
The way I setup the memory agent was to look for an
instruction_typefield that could be anything and its up the individual agent to care. That leaves the door open for more than just "example" or "feedback" if an agent wants to. If it is missing I fall back on example as that's been the assumption so far. We could elevate that to the base_agent API and just leave it up to implementers to ignore if they don't care.eharpste commentedon Dec 7, 2020
Or do you mean the front end side needs something for this?