Continuum Feature AI

AI/ML nodes for fine-tuning LLMs inside your Project Continuum workflows

---

🌐 Part of Project Continuum

This is the AI/ML feature repository for Project Continuum — a distributed, crash-proof workflow execution platform. It provides nodes for training and fine-tuning large language models directly inside your visual workflows.

---

🔥 What Is This

A standalone Gradle project containing AI/ML workflow nodes. Currently features the LLM Trainer (Unsloth) node — fine-tune large language models using LoRA (Low-Rank Adaptation) with Unsloth acceleration, right inside your workflow graph.

Ships as a Spring Boot worker with an auto-managed Python virtual environment for ML execution.

---

🧪 Included Nodes

LLM Trainer (Unsloth)

Fine-tune Large Language Models using LoRA with Unsloth acceleration.

Input	Parquet table with instruction + response columns
Output	Model info — path to LoRA adapter weights, base model, training config
Category	Machine Learning, LLM Training

Supported Base Models:

Provider	Models
Unsloth (fastest)	Phi-4, Phi-4-mini-instruct, Mistral 7B, Llama 3/3.1/3.2, Gemma 2, Qwen 2.5
Microsoft	Phi-2, Phi-3-mini-4k-instruct
Meta	Llama-2-7b, Llama-2-7b-chat
Google	Gemma 2B, Gemma 7B
Qwen	Qwen2-7B, Qwen2-7B-Instruct
TII	Falcon 7B, Falcon 7B-instruct
Custom	Any HuggingFace causal language model

Configurable Parameters:

Group	Parameters
Model	Base model (HuggingFace ID), HuggingFace token for gated models
Data	Input column, output column, system prompt
Training	Epochs, batch size, learning rate, max sequence length, warmup steps, weight decay, gradient accumulation
LoRA	Rank (r), alpha, dropout
Advanced	4-bit quantization, random seed, save steps, logging steps, Parquet batch size

Key Features:

• Unsloth acceleration on Linux + CUDA (2x faster, 60% less memory)
• Falls back to standard HuggingFace transformers on other platforms
• 4-bit quantization for reduced memory usage
• Real-time training progress streaming via Kafka
• Auto-managed Python virtual environment

---

🐍 Python Environment

The Unsloth node executes training via a Python virtual environment that is automatically created at startup if missing.

Setting	Default
com.continuum.feature.ai.unsloth-trainer.venv-path	~/.continuum/unsloth-env
com.continuum.feature.ai.unsloth-trainer.cache-storage-path	./.continuum-cache/workflow-data

Required Python packages (auto-installed): pyarrow, pandas, datasets, torch, transformers, peft, trl, accelerate, hf_transfer, sentencepiece, protobuf, bitsandbytes, unsloth (Linux + CUDA only).

---

📦 Dependencies

Shared libraries from Continuum via GitHub Packages:

Dependency	Purpose
continuum-commons:0.0.1	Base node model, data types, Parquet/S3 utilities
continuum-worker-springboot-starter:0.0.1	Worker framework — registers nodes with Temporal

---

🚀 Quick Start

Prerequisites

• JDK 21 — Eclipse Temurin recommended
• Python 3.10+ — For the Unsloth training environment
• Docker & Docker Compose — For local infrastructure
• GitHub PAT with read:packages scope
• (Optional) CUDA GPU — For Unsloth acceleration

Set environment variables:

export GITHUB_USERNAME=your-github-username
export GITHUB_TOKEN=ghp_your-personal-access-token

Run

# Start infrastructure (Temporal, Kafka, MinIO, API Server, Message Bridge)
cd docker && docker compose up -d

# Build
./gradlew build

# Start the AI worker (auto-creates Python venv on first run)
./gradlew :worker:bootRun

---

📁 Project Structure

continuum-feature-ai/
├── features/
│   └── continuum-feature-unsloth/            # Unsloth LLM trainer node
│       ├── build.gradle.kts                  # Depends on continuum-commons
│       └── src/main/kotlin/.../
│           ├── AutoConfigure.kt              # Spring auto-configuration
│           ├── node/
│           │   └── UnslothTrainerNodeModel.kt
│           └── python/
│               └── PythonEnvironmentManager.kt
├── worker/                                   # Spring Boot worker application
│   ├── build.gradle.kts                      # Depends on starter + unsloth feature
│   └── src/main/
│       ├── kotlin/.../App.kt
│       └── resources/application.yaml
├── docker/                                   # Local development infrastructure
│   └── docker-compose.yml
├── settings.gradle.kts
├── gradle.properties
└── README.md

---

🗺️ Roadmap

• LLM fine-tuning with Unsloth + LoRA
• Auto-managed Python virtual environment
• 4-bit quantization support
• Inference node — run inference against fine-tuned or base models
• Model evaluation node — automated benchmarking
• Multi-GPU training support
• More model architectures (vision, embedding)

---

🔗 Related Repositories

Repository	Description
Continuum	Core backend — API server, worker framework, shared libraries
continuum-workbench	Browser IDE — Eclipse Theia + React Flow workflow editor
continuum-feature-base	Base analytics nodes — data transforms, REST, scripting, anomaly detection
continuum-feature-ai (this repo)	AI/ML nodes — LLM fine-tuning with Unsloth + LoRA
continuum-feature-template	Template — scaffold your own custom worker with nodes