One
One
These are different versions of the OpenChat release. The `OpenChat, OpenChat8192` were trained solely on GPT-4 answers from ShareGPT (`openchat_sharegpt4_dataset`). `OpenChatV3.1, OpenChatV2-W, OpenChatV2` were trained on all ShareGPT data (including...
> Sorry to trouble you again. > > I'm wonder what is updated in `OpenChatV3.2 SUPER` which boosts the score of MT-Bench. Basically hyperparameter tuning :(
> > These are different versions of the OpenChat release. > > The `OpenChat, OpenChat8192` were trained solely on GPT-4 answers from ShareGPT (`openchat_sharegpt4_dataset`). > > `OpenChatV3.1, OpenChatV2-W, OpenChatV2` were...
Hi @huazhenliu We've tried Llama 2 13B, the performance is worse than Mistral 7B, so we've chosen Mistral 7B as the base model. - For your second question, it's OK....
This is expected as Gemma uses tied embeddings, so the `output.weight` should be the same as input embeddings. Could you please open up an issue in llama.cpp repo?
Hi @zestaken, can you provide more information about your local model setup?
Are you talking about the online demo? The parameters of the online demo are as follows 1. We set temperature 0.5, top_p 1, max_tokens 768, with any other sampling parameters...
Have you installed CUDA toolkit? If not, try installing CUDA 12.1
Try creating a new `conda` environment and install ochat?
Here is a beta version wrapping the official openchat server in this repo. We're working on a stable release and corresponding instructions. https://hub.docker.com/r/ochat/openchat-server