4 ms·
Without any changes, I've had great results with openhermes 7b chat. It covers 90% of my GPT-4 usecases, and runs fast. Highly recommend.
by keyle 3y ago
Without any changes, I've had great results with openhermes 7b chat. It covers 90% of my GPT-4 usecases, and runs fast. Highly recommend.
- brucethemoose2 3y agoYi 200K finetunes cover a lot for me. Its not just smart, but the ability to just dump a huge context on it and get something coherent (after a few retries maybe) is really cool.
- MacsHeadroom 3y agoWhat do you run it on?
- brucethemoose2 3y agoA single 3090. I can fit 40K-77K context depending on the severity of the quantization.