kaggle notebook, free of charge, 34b model

[wiiiktor]

https://www.kaggle.com/wiiiktor/kaggle-with-34b-model

It is completely free and works with a high quality 34b-parameter model https://huggingface.co/01-ai/Yi-34B-Chat that has better MMLU scores than Llama-2-70B-chat model (you can find all the scores on their huggingface/github pages). Works rather slow, 1 token per 2 seconds, but maybe the 4bit version (I did not load it yet) is faster. Example of the model generation below (generation is cut to a limit of 77 tokens, as this was just a test). Happy gui-dancing! :-)