Magnet

<p align="center"> <img height="300" width="300" src="./magnet.png"> <br> <h1 align="center">magnet</h1> <h3 align="center"><a href="https://prismadic.github.io/magnet/">📖 docs</a> | 💻 <a href="https://github.com/Prismadic/magnet/tree/main/examples">examples</a> | 📓 <a href="https://prismadic.substack.com">substack</a></h3> <p align="center">the small distributed language model toolkit</p> <p align="center"><i>⚡️ fine-tune state-of-the-art LLMs anywhere, rapidly ⚡️</i></p> <div align="center"> </p>

</div> </p> <img src='./divider.png' style="width:100%;height:5px;">

🧬 Installation

pip install llm-magnet

or

python3 setup.py install

<img src='./divider.png' style="width:100%;height:5px;">

🎉 usage

check out the example notebooks

<small>a snippet to get you started</small>

from magnet.base import Magnet
from magnet.base import EmbeddedMagnet

cluster = EmbeddedMagnet()
cluster.start()
magnet = cluster.create_magnet()
await magnet.align()

config = {
    "host": "127.0.0.1",
    "credentials": None,
    "domain": None,
    "name": "my_stream",
    "category": "my_category",
    "kv_name": "my_kv",
    "session": "my_session",
    "os_name": "my_object_store",
    "index": {
        "milvus_uri": "127.0.0.1",
        "milvus_port": 19530,
        "milvus_user": "test",
        "milvus_password": "test",
        "dimension": 1024,
        "model": "BAAI/bge-large-en-v1.5",
        "name": "test",
        "options": {
            'metric_type': 'COSINE',
            'index_type':'HNSW',
            'params': {
                "efConstruction": 40
                , "M": 48
            }
        }
    }
}

magnet = Magnet(config)
await magnet.align()

<img src='./divider.png' style="width:100%;height:5px;">

🔮 features

<center> <img src="./clustered_bidirectional.png" style="width:50%;transform: rotate(90deg);margin-top:200px;" align="right"> </center>

• ⚡️ It's Fast
  • <small>fast on consumer hardware</small>
  • <small>very fast on Apple Silicon</small>
  • <small>extremely fast on ROCm/CUDA</small>
• 🫵 Automatic or your way
  • <small>rely on established transformer patterns to let magnet do the work</small>
  • <small>keep your existing data processing functions, bring them to magnet!</small>
• 🛰️ 100% Distributed
  • <small>processing, embedding, storage, retrieval, querying, or inference from anywhere</small>
  • <small>as much or as little compute as you need</small>
• 🧮 Choose Inference Method
  • <small>HuggingFace</small>
  • <small>vLLM node</small>
  • <small>GPU</small>
  • <small>mlx</small>
• 🌎 Huge Volumes
  • <small>handle gigantic amounts of data inexpensively</small>
  • <small>fault-tolerant by design</small>
  • <small>decentralized workloads</small>
• 🔐 Secure
  • <small>JWT</small>
  • <small>Basic</small>
• 🪵 World-Class Comprehension
  • <small>magnet optionally logs its own code as it's executed (yes, really)</small>
  • <small>build a self-aware system and allow it to learn from itself</small>
  • <small>emojis are the future</small>

<img src='./divider.png' style="width:100%;height:5px;">

🧲 why

• build a distributed LLM research node with any hardware, from Rasbperry Pi to the expensive cloud
• Apple silicon first-class citizen with mlx
• embed & index to vector db with milvus
• distributed processing with NATS
• upload to S3
• ideal cyberpunk vision of LLM power users in vectorspace

<img src='./divider.png' style="width:100%;height:5px;">