But what local llm model I can use?

For running LLM model locally you'll require a good GPU machine.

There is a LLM leaderboard on HF that ranks open-source LLM. Find more here: https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard

LlamaIndex has also created a benchmark for LLMs. Find it here: https://docs.llamaindex.ai/en/stable/module_guides/models/llms.html#llm-compatibility-tracking

I use HuggingFaceH4/zephyr-7b-beta with some success.

See https://colab.research.google.com/drive/1UoPcoiA5EOBghxWKWduQhChliMHxla7U