it looks like the energy is presently focused around hooking the salesforce codegen 6b model (said to be a strong model, said to run on one modern gpu) with carperai’s trlx project (reinforcement learning with human feedback).
huu replies ro the question about @home but maybe glanced over the question about directing dev effort. anyway it’s cool it’s moving forward. https://discord.com/channels/823813159592001537/1051442019035254784/1057767521748537355 is @home an existing inference structure? i reskimmed trlx and see it is likely not networked training. if i start on hivemind work, can you guys send interested devs my way if they ask about it?
