geniusrise 0.0.7

An LLM framework

🧠 Geniusrise

---

About

Geniusrise is a modular, loosely-coupled AgentOps / MLOps framework designed for the era of Large Language Models, offering flexibility, inclusivity, and standardization in designing networks of AI agents.

It seamlessly integrates tasks, state management, data handling, and model versioning, all while supporting diverse infrastructures and user expertise levels. With its plug-and-play architecture, Geniusrise empowers teams to build, share, and deploy AI agent workflows across various platforms efficiently.

TLDR 🙄

1. Install geniusrise

pip install geniusrise
pip install geniusrise-huggingface

2. Create genius.yaml

version: '1'
bolts:
    HuggingFaceInstructionTuningBolt:
        name: 'hf-fine-tune-my-shit'
        method: fine_tune
        args:
            model_name: bert-base-uncased
            tokenizer_name: bert-base-uncased
            batches: 2
            hf_repo_id: my/repo
            token: 'hf_woohoo'
            commit_message: say hello to genius!
        input:
            type: batch
            args:
                bucket: geniusrise-test
                folder: my-shit
        output:
            type: batch
            args:
                bucket: geniusrise-test
                folder: my-model
        deploy:
            type: 'k8s'
            args:
                cluster_name: my-cluster
                namespace: geniusrise-huggingface
                labels: {'needs': 'gpu'}
                cpu: 16
                memory: 50G
                storage: 250Gb
                gpu: 1

3. Copy data to s3

cat > data.jsonl <<- EOM
{"instruction": "instruction1", "output":"output1"}
{"instruction": "instruction2", "output":"output2"}
EOM

aws s3 cp data.jsonl s3://geniusrise-test/my-shit/

4. Fine tune

genius --yaml genius.yaml deploy

🙄 This was not even crux of the iceberg. Please see docs.

Links

• Website: geniusrise.ai
• Docs: docs.geniusrise.ai
• Hub: geniusrise.com [coming soon]