TabbyML

How to install for `Apple Silicon` via `homebrew`

Ref: https://tabby.tabbyml.com/docs/installation/apple

# Install homebrew
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install tabby
brew install tabbyml/tabby/tabby

# Serve completion and chat with some model https://tabby.tabbyml.com/docs/models/
tabby serve --device metal --model TabbyML/DeepseekCoder-6.7B --chat-model TabbyML/Mistral-7B

# Or
tabby serve --device metal --model TabbyML/Mistral-7B --chat-model TabbyML/Mistral-7B

# Config to indexing some repo
cat << 'EOF' > ~/.tabby/config.toml
[[repositories]]
git_url = "https://github.com/rust-lang/book.git"

[[repositories]]
git_url = "https://github.com/rust-lang/rust-by-example.git"
EOF

# Indexing now
tabby scheduler --now

How to add more repos manually

Open config file

open ~/.tabby/config.toml

Then add some repo

[[repositories]]
git_url = "https://github.com/DioxusLabs/dioxus.git"

Finally force indexing

tabby scheduler --now

RAM used

--model TabbyML/DeepseekCoder-6.7B // RAM used 2.1GB
--chat-model TabbyML/Mistral-7B // RAM used 2.6GB

How to run `TabbyML` via `Windows`

Install Docker via WSL

sudo docker run -it --gpus all -p 8080:8080 -v $HOME/.tabby:/data tabbyml/tabby serve --model TabbyML/StarCoder-1B --device cuda

And you will get an error.

docker: Error response from daemon: could not select device driver "" with capabilities: [[gpu]].

The fix

distribution=$(. /etc/os-release;echo $ID$VERSION_ID)
curl -s -L https://nvidia.github.io/nvidia-docker/gpgkey | sudo apt-key add -
curl -s -L https://nvidia.github.io/nvidia-docker/$distribution/nvidia-docker.list | sudo tee /etc/apt/sources.list.d/nvidia-docker.list

sudo apt-get update && sudo apt-get install -y nvidia-container-toolkit
sudo systemctl restart docker

How to select model

Optional code completion model use TabbyML/CodeLlama-7B

sudo docker run -it --gpus all -p 8080:8080 -v $HOME/.tabby:/data tabbyml/tabby serve --model TabbyML/CodeLlama-7B --device cuda

Optional code completion model use TabbyML/Mistral-7B

sudo docker run -it --gpus all -p 8080:8080 -v $HOME/.tabby:/data tabbyml/tabby serve --model TabbyML/Mistral-7B --device cuda

Optional chat model use TabbyML/Mistral-7B

⚠️ I can't make this one work, it's just crash and exit. 🤔

sudo docker run -it --gpus all -p 8080:8080 -v $HOME/.tabby:/data tabbyml/tabby serve --model TabbyML/StarCoder-1B --chat-model TabbyML/Mistral-7B --device cuda

How to build and run docker locally to match your cuda version e.g. 12.3.0

docker build --build-arg CUDA_VERSION=12.3.0 -t tabby_cuda12_3 .
docker run -it --gpus all -p 8080:8080 -v $HOME/.tabby:/data tabby_cuda12_3 serve --device cuda --model TabbyML/StarCoder-1B --chat-model TabbyML/Mistral-7B

How to get code completion = index from target repos

Optional schedule now

Refer to https://tabby.tabbyml.com/blog/2023/10/16/repository-context-for-code-completion/
```
sudo docker run -v $HOME/.tabby:/data tabbyml/tabby scheduler --now
```

Or schedule via running docker

sudo docker ps -a | grep tabby | awk '{print $1}' | xargs sudo docker exec -it $1 sh -c "/opt/tabby/bin/tabby scheduler --now"

How to request the `TabbyML` services from other machine to Windows WSL2

See your host info
```
wsl hostname -I
ipconfig
```
Open Windows Firewall→Advanced Settings and create new Inbound Rules for Your local IP4 (e.g. 192.168.1.33) that allow port 8080.

Then forward port 8080 to WLS2

netsh interface portproxy add v4tov4 listenaddress=192.168.1.33 listenport=8080 connectaddress=127.0.0.1 connectport=8080

Open in your browser.
```
open http://192.168.1.33:8080
```

Ideas

CLI Lazy git.
CLI Auto fix after compile.
CLI Model selection.
CLI Configurable repos.
CLI Indexing manual trigger.
Query Include/Exclude repos for faster query.
Query Include/Exclude language for faster query.
Embedding Code in comment?, PDF, Table, Image.

How to dev tabby

# Setup
git clone --recurse-submodules https://github.com/TabbyML/tabby
cd tabby

# macos
brew install protobuf
brew install cmake

# Update
git pull
git submodule update --init --recursive

How to build

ref: https://github.com/rust-lang/rust/issues/117976

## Workaround for Rust 1.17.4
rustup default nightly

## Build release
cargo build --release

Gist of Rust