facebookresearch
Doc improvement requests
Please comment with docs you think should be updated, added, for clarified.
I want to write my own bot by inheriting World, Agent, ... but feel lost on current instruction. I hope the team could create a simple guideline to create a custom task and integrate it to, for example, Facebook messenger.
It would be nice to have better installation instructions and specific package dependencies listed in the hallucination project. Right now it's an ocean of upgrading and downgrading packages until something works.
It would be nice to have better installation instructions and specific package dependencies listed in the hallucination project. Right now it's an ocean of upgrading and downgrading packages until something works.
Thanks for the feedback!
I would certainly appreciate it if there was a document detailing the Blenderbot2 specific command line arguments and parameters.
A lot of them are fairly straightforward, but there are also quite a few that, while they look useful, I can’t find info on what they do. Just basic info for someone just starting out using this architecture. Thanks!
Hi @stephenroller, Thank you for taking this initiative. Since there is no complete paper on BlenderBot 2, it makes it hard for us to understand the working of code and how they are combined together. It would be helpful if we can create a document/flowchart explaining the workflow and the code.
I will be glad to contribute to that process with what I know so far. Thanks!
@atharvjairath Ok that explains the lack of info. In that case I will be looking at the code the best I can, and trying to work through some of it. If you or someone else can start that document and maybe figure out a layout —I’m not very good at organizing things— I can add what I find. I’ll be keeping notes until then and trying to work out descriptions for the parameters.
@Jcwscience, That sounds great. I am a student, therefore, could only understand some part of it, but I am willing to understand it whole. I suggest we can take this to discussions, and create a Google Docs for easy editing, where we can decide the layout, notes, parameters, architecture, etc, to use and explain the workflow of BlenderBot2. After that, I can convert that to README and push it here I also hope that many others would join and fill the missing pieces in our understanding.
@stephenroller @klshuster @mojtaba-komeili Any suggestions/advice would be appreciated.
@atharvjairath I have been making some progress, however I am running into a wall. I know that the model encodes and saves both previous conversations and I think new info from the search server, but I can’t figure out where it saves session info. Or a way to start it other then the parlai interactive command for that matter.
@Jcwscience I am not sure about that too, what I read/heard is that it stores all the information in batches, and access them if it matches the context. I also read in some issues that, some convo from user1 can be continued by user2. I have no information about how to continue the session again. I guess right now it's working in one-to-one method and not one-to-many.
Hi there, I would like to train my own model using BlenderBot but having hard times following the tutorials on the website. Are there any clear step-by-step guide?
Hi, I couldn't find any documentation about the computational needs for every model? It would be quite useful!
All the scripts Speeding up Training guide at https://www.parl.ai/docs/tutorial_fast.html# has an incorrect syntax. The --task in the first line gives argument task --t expects 1 argument.
