alex/discord-aidolls
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Switch Model Command (#126)

Browse Source
This commit is contained in:
Kevin Dang
2024-10-12 22:03:31 -07:00
committed by GitHub
parent 5d02800c3f
commit 9f61f6bc6c
22 changed files with 334 additions and 376 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
docs/setup-docker.md
View File

@@ -2,7 +2,9 @@
* Follow this guide to setup [Docker](https://www.digitalocean.com/community/tutorials/how-to-install-and-use-docker-on-ubuntu-20-04)
* If on Windows, download [Docker Desktop](https://docs.docker.com/desktop/install/windows-install/) to get the docker engine.
* Please also install [Docker Compose](https://docs.docker.com/compose/install/linux/) for easy running. If not, there are [scripts](#manual-run-with-docker) to set everything up.
* **IMPORTANT NOTE**: Currently, it seems like wsl does not like Nvidia Container Toolkit. It will work initially then reset it for some odd reason. For now, it is advised to use an actually Linux machine to run using Docker. If you do not care about utilizing your GPU or don't even have a Nvidia GPU then disregard this.
> [!IMPORTANT]
> Currently, it seems like wsl does not like Nvidia Container Toolkit. It will work initially then reset it for some odd reason. For now, it is advised to use an actually Linux machine to run using Docker. If you do not care about utilizing your GPU or don't even have a Nvidia GPU then disregard this.
## Nvidia Container Toolkit Setup
### Installation with Apt
@@ -48,8 +50,6 @@ sudo systemctl restart docker
* `SUBNET_ADDRESS = 172.18.0.0`
* Don't understand any of this? watch a Networking video to understand subnetting.
* You also need all environment variables shown in [`.env.sample`](../.env.sample)
* You will need a model in the container for this to work properly, on Docker Desktop go to the `Containers` tab, select the `ollama` container, and select `Exec` to run as root on your container. Now, run `ollama pull [model name]` to get your model.
* For Linux Servers, you need another shell to pull the model, or if you run `docker compose build && docker compose up -d`, then it will run in the background to keep your shell. Run `docker exec -it ollama bash` to get into the container and run the samme pull command above.
* Otherwise, there is no need to install any npm packages for this, you just need to run `npm run start` to pull the containers and spin them up.
* For cleaning up on Linux (or Windows), run the following commands:
* `docker compose stop`

7
docs/setup-local.md
View File

@@ -11,14 +11,17 @@
* You can now interact with the model you just ran (it might take a second to startup).
* Response time varies with processing power!
> [!NOTE]
> You can now pull models directly from the Discord client using `/pull-model <model-name>` or `/switch-model <model-name>`. They must exist from your local model library or from the [Ollama Model Library](https://ollama.com/library)
## To Run Locally (without Docker)
* Run `npm install` to install the npm packages.
* Ensure that your [.env](../.env.sample) file's `OLLAMA_IP` is `127.0.0.1` to work properly.
* You only need your `CLIENT_TOKEN`, `MODEL`, `OLLAMA_IP`, `OLLAMA_PORT`.
* You only need your `CLIENT_TOKEN`, `OLLAMA_IP`, `OLLAMA_PORT`.
* The ollama ip and port should just use it's defaults by nature. If not, utilize `OLLAMA_IP = 127.0.0.1` and `OLLAMA_PORT = 11434`.
* Now, you can run the bot by running `npm run client` which will build and run the decompiled typescript and run the setup for ollama.
* **IMPORTANT**: This must be ran in the wsl/Linux instance to work properly! Using Command Prompt/Powershell/Git Bash/etc. will not work on Windows (at least in my experience).
* Refer to the [resources](../README.md#resources) on what node version to use.
* If you are using wsl, open up a separate terminal/shell to startup the ollama service. Again, if you are running an older ollama, you must run `ollama serve` in that shell.
* If you are on an actual Linux machine/VM there is no need for another terminal (unless you have an older ollama version).
* If you do not have a model, you will need to run `ollama pull [model name]` in a separate terminal to get it.
* If you do not have a model, you **can optionally** run `ollama pull [model name]` in wsl prior to application start. You are not required as it can be pulled from the Discord client.