Closed Ananderz closed 1 year ago
I was able to get the model to work through GGML. Only tricky part was having to find GGML version of the Falcon model on Hugging Face, rather than just using the default one from the leaderboards.
Support for the new GGML Falcon models was added recently in ctransformers 0.2.12 Make sure you have installed the latest version:
pip install 'ctransformers>=0.2.12'
You can find the falcon GGML models here. Example:
Normal HF and GPTQ versions should also work.
I know the falcon models are a little different and that they might not work with Chatdocs. Have anyone tried and got it to run. Which model did you use? GGML, GPTQ or maybe the standard models?