constellaxion 0.4.3

The constellaXion CLI for managing your laboratory database

constellaXion CLI: Automated LLM Deployments for your private cloud

---

The fastest way to Train, Deploy and Serve Open Source Language Models to your Private Cloud

⚡️ Features

Configure and access the most popular open source LLMs with a few simple commands

• 📄 YAML-based configuration

• ⚙️ Fine-tune LLMs with your own data and cloud resources

• 🚀 Deploy models to your private cloud

• 🤖 Serve your models with ease

• 💬 Prompt your models with ease

📚 The Stack

ConstellaXion leverages industry-leading technologies to provide a seamless LLM deployment experience. Deploy to AWS or GCP, serve models efficiently with vLLM, access the latest models from Hugging Face, and utilize DJL's powerful serving capabilities.

🔧 Quick Start

Installation

Install the package:

pip install constellaxion

For Windows users: this package may compile dependencies (e.g., numpy) from source. Please ensure you have the Microsoft C++ Build Tools installed: https://visualstudio.microsoft.com/visual-cpp-build-tools/

Alternatively, install with prebuilt binaries:

pip install --prefer-binary constellaxion

YAML Configuration Format

Create a model.yaml file to describe your project. Example:

model:
  id: my-llm
  base: tiiuae/falcon-7b-instruct

dataset:
  train: ./train.csv # Path to your training data.
  val: ./val.csv # Path to your validation data
  test: ./test.csv # Path to your test data

training:
  epochs: 1
  batch_size: 4

deploy:
  gcp:
    project_id: my-gcp-project
    region: europe-west2

model (required): The ID of the model you want to use.

base (required): The base model you want to use (HuggingFace path).

dataset (required for finetuning): The path to your training, validation, and test data.

• All 3 datasets must be provided as CSV files with the following columns: prompt, response.

training (required for finetuning): The number of epochs and batch size for training.

deploy (required for deployment): The Deployment target. Currently only GCP is supported.

• gcp:
  • project_id: The GCP project ID.
  • region: The GCP region to deploy the model to.

Usage

Initialize your project:

constellaXion init

View your current model configuration:

constallaXion model view

Deploy a foundation model without finetuning:

constellaXion model deploy

Run fine-tuninig job (Dataset and training configs required):

constallaXion model train

Serve a fine-tuned model:

constallaXion model serve

Prompt model:

constallaXion model prompt

Example Workflow

Create model.yaml file: to fit your use case

constellaXion init

Print the current training and serving configuration from your YAML file:

constellaXion model view

Run training job:

constellaXion model train
constellaXion model serve
constellaXion model prompt