Models

[wiseman-timelord]

The model to use is Vicuna, as it has best objective information score, slightly better than even GPT4-X-Alpaca. Vicuna is now at v1.1, and I can verify "ggml-vicuna-13b-1.1-q4_1.bin" from "https://huggingface.co/eachadea/ggml-vicuna-13b-1.1" is compatible.
Some versions of GPT4-X-Alpaca also work, specifically I know "gpt4-x-alpaca-13b-native-ggml-model-q4_0" from "https://huggingface.co/Bradarr/gpt4-x-alpaca-13b-native-ggml-model-q4_0" works, but, didn't take too long over testing, for reasons stated.

The 1.1 Vicuna seems better than the 1.0 for AutoLLM, from short review.