From a837ea48e017289a21a9574b0fe862f541874a14 Mon Sep 17 00:00:00 2001 From: Nathan Sarrazin Date: Tue, 21 Mar 2023 07:54:57 +0100 Subject: [PATCH] Updated readme according to the latest changes, hopefully fixes #1 --- README.md | 10 ++++++++-- 1 file changed, 8 insertions(+), 2 deletions(-) diff --git a/README.md b/README.md index 9ade5a7..6a9bef1 100644 --- a/README.md +++ b/README.md @@ -20,16 +20,22 @@ Setting up Serge is very easy. Start by cloning the repo: git clone git@github.com:nsarrazin/serge.git ``` -Then put your weights in the `models` folder. If you don't have them you can download them using the following magnet links: +Then put your weights in the `api/weights` folder. If you don't have them you can download them using the following magnet links: - [ggml-alpaca-7b-q4.bin](https://maglit.me/corotlesque) - [ggml-alpaca-13b-q4.bin](https://maglit.me/nonchoodithvness) They are currently the only two models supported. I'm working on expanding support to all the models supported by `llama.cpp`. +You will also need to put the `tokenizer.model` for LLaMa in that folder. [You can grab it here](https://huggingface.co/decapoda-research/llama-7b-hf/blob/main/tokenizer.model). + +### Model conversion + Note: `llama.cpp` [recently underwent some change](https://github.com/ggerganov/llama.cpp/issues/324#issuecomment-1476227818) that requires model weights to be converted to a new format. Serge picks this up automatically on startup, and will convert your weights to the new format if needed. The old weights will be renamed to `*.bin.old` and the new weights will be named `*.bin`. -Then, you can start the project by running: +### Starting the project + +Finally you can start the project by running: ``` cp .env.sample .env