So I have recently gone down the rabbit hole of cancelling my ChatGPT subscription and now just use OpenHermes2.5-Mistral-7B. I’ve learned about the different benchmarks and how they compare and I understand how to read the HuggingFace LLM leaderboard and download any other model I might like to try.
What I struggle to understand is the meaning of the naming conventions. Mistral seems to clearly be better than LLAMA2 from what I have read and I understand the differences of 7B, 13B, etc etc.
Can someone explain the additional prefixes of Hermes, OpenHermes, NeuralChat, etc.
Tldr; What is the difference between Dolphin-Mistral and OpenHermes-Mistral. I’m guessing one is the dataset and the other is how it was trained?
Mistral and Llama2 (and Llama) are foundation models, meaning they actually trained all the weights given. Almost anything worth using is a derivative of these 3 foundation models. They are really expensive to train.
Just about everything else is a Lora fine tune on top of one of them. Fine tunes only change a small fraction of the weights, like 1%. Functionally speaking, the important part of these is the additional data they were trained on, and that training can be done on any underlying model.
So Open hermes is a Lora tuning on top of mistral, and is some opensource offshoot of nous hermes, which is an instruction dataset for giving good smart answers (or something) in a given instruction format.