Hosted version of Monty

General
1

Hi,

“This is more of a comment than a question” — looking around at the existing AI models, they all seem to have some hosted version for consumption (with language, granted, which doesn’t seem to be a current focus for Monty). Still: how do you see/envision the hosted aspect of TBP?

Thank you,
Kate

1 Like
2

Hey there Kate,

Great question. I think this is still an open conversation within the TBP team. Currently, they run their models on AWS with 16 cores and quite a bit of memory (not sure on the CPU speed). They’re not currently using any kind of GPU acceleration.

Monty’s first target demographic, as far as I can tell, is in edge computing, specifically, in robotics. Traditionally, these platforms are pretty resource constrained and often have limited network connectivity. So the kinds of scalability we see made available through cloud infrastructure is less accessible/useful.

For what its worth though, we’ve had a lot of people here test it on different hardware. A community favorite is the raspberry pi. Personally, I’ve ran my instance on an i7 7700 (older CPU at this point) with 16 gigs of RAM. I guess what I’m trying to say is that you can run Monty on some pretty cheap hardware. Additionally, you could always spin up a VM in the cloud and deploy there, if you so desired.

1 Like
3

Welcome aboard, @radkat! One issue that @HumbleTraveller finessed was how the sensor data would get to a copy of Monty residing in a cloud-based VM. Here are some possibilities:

  • A local machine (e.g., a RasPi) could host Sensor and/or Effector Modules. It would use the Cortical Message Protocol (really, data format) and some networking juju to work with the cloud-based instance of Monty.

  • A cloud-based instance of Monty could exchange CMP messages with other cloud residents. For example, there could be a Sensor Module that gathers data from telemetry providers, traces, etc.

  • A cloud-based sensor module could interact with a video provider (e.g., YouTube). Monty might use this to watch and listen to presentations, then generate accessible slide descriptions, transcripts, etc.

Of course, all of this is still quite speculative…

1 Like
4

Hi @radkat, welcome!

I’m curious: if a hosted Monty was available to you, what would that do for you?

1 Like
5

Great question. I guess I’m after two things at once: 1. be Monty’s user (with lowest bar entry possible) and 2. find a way to tinker with it within my possibilities.

For the first one, it’d be ideal if either I could easily spin it up myself (e.g., with instructions on how to get a docker container, etc.) or if there was a playground version for me to use.

And here comes the second bullet: after that point, I’d be still figuring out what exactly to do with it b/c I’m not from the robotics world and I’d prefer some language-based interaction — but, say, I do have home cameras (or could try to feed it my photos) – but right now with just python code base I don’t know how to even start holding it.

Basically, I’m really into the idea behind TBP but as a non-sophisticated gen AI consumer I’ve no clue how to start being engaged.

1 Like
6

I don’t know how well the current Monty implementation would fit into a Docker container, but this could be a convenient and useful way to distribute WIP snapshots. Perhaps “someone” :slight_smile: should look into setting up a GitHub Action (or whatever) for this.

7

Thank you, @radkat. That’s very helpful and gives us some ideas about how to make Monty more accessible.

8

Hi @radkat thanks for the question and explaining what you are looking for and thanks to everyone who already replied!
Just a few more thoughts from my side to add in case you are still interested in an easy way to get started with Monty.

  1. The nice thing about Monty is that it is pretty lightweight (at least at the current scale we usually use it at) and can easily run on any laptop without requiring any GPU or larger cloud infrastructure. Like @HumbleTraveller already said, you don’t need expensive hardware to run some first experiments with Monty.
  2. The easiest entry point to Monty would be our tutorials. You’ll have to go through the setup instructions first and then the tutorials take you step-by-step through some example experiments with Monty, including code and explanations.
  3. Also, if you want to get a general overview of the project and how the implementation works, you could start with our youtube videos. This one from our launch symposium may be a good starting point to get a feel for what Monty can and cannot do and how it works.

Hope this helps!
-Viviane

1 Like