Kurt Shuster
Kurt Shuster
think this all looks good on my end but will defer to @moyapchen for final approval
yes! could you please update the cache keys one more time?
hey @davides, is this ready for review?
Hey @davides, should we close this or are you still working on it?
can you please provide your command for running this agent? something's off if your decision modules are generating incorrect tokens (e.g., `Example 0, search_decision_agent: My name is jessica` is nonsense)
is your fine-tune data appropriately formatted? BB3 makes use of specific control tokens that determine which module to output. if your data does not have the appropriate control tokens you...
I would take a look at the [bb3 agent README](https://github.com/facebookresearch/ParlAI/tree/main/projects/bb3/agents) for how to format your tasks appropriately. Indeed if the wrong control tokens are used at training time, you may...
> In terms of multi-task training with MSC and ConvAI2, did anything change in the formatting of the training tasks when training with BB3? As I said, please see the...
In addition, I'd recommend creating a custom task file that subclasses the jsonfile agent, and overrides the [`_get_ep_from_turns`](https://github.com/facebookresearch/ParlAI/blob/5f41ba772d171f1e9f96ea1c9d168ad12a3bb4c9/parlai/core/teachers.py#L1729) function to parse your context accordingly
> As suggested by @klshuster , I can add more context apart from text and labels for creating the task? you can access those `context` values (and place them in...