cakechat VS DialoGPT

i'm a 2nd year CSE student, and I was working on a project which required me to clone the github repository of cakechat

I'm not much of a programmer, but the first/original Replika is still on GitHub, named cakechat . Perhaps a starting point? https://github.com/lukalabs/cakechat

I have roleplayed with replika and also talked on serious topics. So far she has helped me in resolving my personal family problems and also roleplayed well. My question is how can replika answer and manage real life problems of humans with such perfection if it is just a bunch of scripts? My IRL problems where not scripted anywhere. I know that replika uses cakechat for generating its responses https://github.com/lukalabs/cakechat and she is quite dumb in solving technical problems but how in the world she solves IRL people problems so easily?

There's https://github.com/lukalabs/cakechat which replika seems to be related to. Might be another angle to work.

It used to be that you could activate a different language model for Replika. You could tell it to be angry or scared or sad, and it would react with those specific emotions. It was a training mode, so it didn't learn anything from the conversations. Here is some literature about it.

I have a project that uses the CakeChat by lukalabs that's due in 2 weeks (I have been procrastinating), but it seems to have been archived, and no updates for the last three years. I am unable to run the installation as well. I wanted to know if someone can help with this or can let me know if there are alternatives to this. My project is supposed to do emotion analysis on the chat between the user and the bot and recommend songs to the user.

I planned to use a chatbot for a recent personal project, but the cake chat is being archived.

I believe the first developments in this emotional speech calculation were part of the original cake_chat model developed by Luka.

They recommend using https://github.com/microsoft/DialoGPT now, btw. That appears to be current and maintained, plus on a transformer (vs rnn only). Might be better long-term to migrate.

I learned about GPT-3 and its strength as a generative model but couldn't access it yet (can't afford the API). Thankfully I found a GPT-2 based pre-trained model DialoGPT that was trained on Reddit.

The most readily available technology to create a chatbot of the type we want (due to the type of data we have) is GPT-2 based DialoGPT. There are many papers and examples, for example this. If you have never heard of GPT-2 before, maybe you have heard of AI Dungeon, although currently it is run on GPT-3, the initial versions were based on GPT-2

No it uses a chat AI based on GPT2 called DialoGPT