Spaces:
Running
You might want to lower the Temperature from 0.95 to 0.85
The responses I get from Open Assistant (https://open-assistant.io/chat) are amazingly good compared to the ones I get here on HuggingChat (https://huggingface.co/chat/).
Is it because HuggingChat has the temperature set really high (0.95) that causes it to ramble like a schizophrenic?
Even when using the Creative preset, it's very very coherent -- almost no rambling, even with a very long conversation including followup questions to the responses.
I get that you guys might want to funnel people to that other site (to help provide feedback for the AI model), but the incoherence of the AI model here gives it a bad impression that might turn people off completely.
I think you're using the same model as their "oasst-sft-6-llama-30b" option, so the only other difference should be the Temperature -- their Creative preset uses 0.85 and all the other settings are identical afaict.
I can confirm that. I'm using it in a Telegram bot and adjusted the temperature to 0.85 and it's now more coherent.