mlipy 0.1.7

Python-based Machine Learning Interface

mlipy

Pure Python-based Machine Learning Interface for multiple engines with multi-modal support.

Python HTTP Server/Client (including WebSocket streaming support) for:

• candle
• llama.cpp
• LangChain

Prerequisites

Debian/Ubuntu

sudo apt update -y
sudo apt install build-essential git curl libssl-dev libffi-dev pkg-config

Rust

1. Using latest system repository:

sudo apt install rustc cargo

2. Install rustup using official instructions:

curl --proto '=https' --tlsv1.2 -sSf https://sh.rustup.rs | sh
source "$HOME/.cargo/env"
rustup default stable

Python

1. Install Python using internal repository:

sudo apt install python3.11 python3.11-dev python3.11-venv

2. Install Python using external repository:

sudo add-apt-repository ppa:deadsnakes/ppa
sudo apt update -y
sudo apt install python3.11 python3.11-dev python3.11-venv

Arch/Manjaro

Rust

1. Using latest system-wide rust/cargo:

sudo pacman -Sy base-devel openssl libffi git rust cargo rust-wasm wasm-bindgen

2. Using latest rustup:

sudo pacman -Sy base-devel openssl libffi git rustup
rustup default stable

macOS

brew update
brew install rustup
rustup default stable

llama.cpp

cd ~
git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
make

candle

cd ~
git clone https://github.com/huggingface/candle.git
cd candle
find candle-examples -type f -exec sed -i 's/println/eprintln/g' {} +
cargo clean
cargo build -r --bins --examples

Run Development Server

Setup virtualenv and install requirements:

git clone https://github.com/mtasic85/mlipy.git
cd mlipy

python3.11 -m venv venv
source venv/bin/activate
pip install poetry
poetry install

Download models:

# NOTE: login in case you need to accept terms and conditions for some models
# huggingface-cli login

# phi
huggingface-cli download microsoft/phi-1_5
huggingface-cli download Open-Orca/oo-phi-1_5
huggingface-cli download lmz/candle-quantized-phi

# stable-lm
huggingface-cli download stabilityai/stablelm-3b-4e1t
huggingface-cli download lmz/candle-stablelm-3b-4e1t

# mistral
huggingface-cli download TheBloke/Mistral-7B-v0.1-GGUF mistral-7b-v0.1.Q4_K_M.gguf
huggingface-cli download TheBloke/Mistral-7B-Instruct-v0.1-GGUF mistral-7b-instruct-v0.1.Q4_K_M.gguf
huggingface-cli download TheBloke/zephyr-7B-beta-GGUF zephyr-7b-beta.Q4_K_M.gguf
huggingface-cli download TheBloke/Yarn-Mistral-7B-128k-GGUF yarn-mistral-7b-128k.Q4_K_M.gguf
huggingface-cli download lmz/candle-mistral

# llama2
# huggingface-cli download meta-llama/Llama-2-7b-hf
huggingface-cli download TheBloke/Orca-2-7B-GGUF orca-2-7b.Q4_K_M.gguf
huggingface-cli download TheBloke/Llama-2-7B-GGUF llama-2-7b.Q4_K_M.gguf
huggingface-cli download TheBloke/Llama-2-7B-Chat-GGUF llama-2-7b-chat.Q4_K_M.gguf
huggingface-cli download TheBloke/Yarn-Llama-2-7B-128K-GGUF yarn-llama-2-7b-128k.Q4_K_M.gguf
huggingface-cli download afrideva/TinyLlama-1.1B-Chat-v0.6-GGUF tinyllama-1.1b-chat-v0.6.q4_k_m.gguf
huggingface-cli download afrideva/TinyLlama-1.1B-intermediate-step-955k-token-2T-GGUF tinyllama-1.1b-intermediate-step-955k-token-2t.q4_k_m.gguf

# stable-lm
huggingface-cli download afrideva/stablelm-3b-4e1t-GGUF stablelm-3b-4e1t.q4_k_m.gguf
huggingface-cli download TheBloke/rocket-3B-GGUF rocket-3b.Q4_K_M.gguf

# code mistral-based
# huggingface-cli download codellama/CodeLlama-7b-Python-hf
huggingface-cli download TheBloke/sqlcoder-7B-GGUF sqlcoder-7b.Q4_K_M.gguf
huggingface-cli download TheBloke/deepseek-coder-1.3b-instruct-GGUF deepseek-coder-1.3b-instruct.Q4_K_M.gguf
huggingface-cli download TheBloke/deepseek-coder-6.7B-instruct-GGUF deepseek-coder-6.7b-instruct.Q4_K_M.gguf
huggingface-cli download TheBloke/tora-code-7B-v1.0-GGUF tora-code-7b-v1.0.Q4_K_M.gguf

Run server:

python -B -m mli.server

Run Examples

python -B examples/sync_demo.py
python -B examples/async_demo.py
python -B examples/langchain_sync_demo.py
python -B examples/langchain_async_demo.py

Run Production Server

python3.11 -m venv venv
source venv/bin/activate
pip install -U mlipy
python -B -m mli.server