Transformers
mpt
Composer
MosaicML
llm-foundry
StreamingDatasets
text-generation-inference
6 papers

Have plan to create a ggml version for mpt-7b-chat?

#4
by LouiSum - opened

As the chat version have the balance when generating content with an instruction. I believe many people want it running locally

So far, AFAIK, nobody makes it. So, We must convert it manually.

  1. Download https://huggingface.co/mosaicml/mpt-7b-chat/tree/main the pretraining model in HF format
  2. Convert HF to GGML using GGML script in repo https://github.com/ggerganov/ggml. We need modify a bit ggml/examples/mpt/convert-h5-to-ggml.py so it can use the pretrained model as we downloaded it in first step.

I know it is not easy enough. Good luck

Oh did I miss one? Then yeah I'll do it

Oh did I miss one? Then yeah I'll do it

So many models, so little time, huh? :D

Sign up or log in to comment