{"id":48100465,"url":"https://github.com/projectcontinuum/continuum-feature-ai","last_synced_at":"2026-04-04T15:42:31.933Z","repository":{"id":342762547,"uuid":"1175041999","full_name":"projectcontinuum/continuum-feature-ai","owner":"projectcontinuum","description":"AI and ML features for continuum","archived":false,"fork":false,"pushed_at":"2026-03-25T07:02:04.000Z","size":160,"stargazers_count":2,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-03-25T14:52:35.315Z","etag":null,"topics":["ai","continuum","continuum-feature","cuda","llm","ml","mlops","pytourch","unsloth"],"latest_commit_sha":null,"homepage":"","language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/projectcontinuum.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":"AGENTS.md","dco":null,"cla":null}},"created_at":"2026-03-07T06:27:04.000Z","updated_at":"2026-03-25T07:02:06.000Z","dependencies_parsed_at":"2026-03-07T15:01:51.807Z","dependency_job_id":null,"html_url":"https://github.com/projectcontinuum/continuum-feature-ai","commit_stats":null,"previous_names":["projectcontinuum/continuum-feature-ai"],"tags_count":0,"template":false,"template_full_name":"projectcontinuum/continuum-feature-template","purl":"pkg:github/projectcontinuum/continuum-feature-ai","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/projectcontinuum%2Fcontinuum-feature-ai","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/projectcontinuum%2Fcontinuum-feature-ai/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/projectcontinuum%2Fcontinuum-feature-ai/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/projectcontinuum%2Fcontinuum-feature-ai/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/projectcontinuum","download_url":"https://codeload.github.com/projectcontinuum/continuum-feature-ai/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/projectcontinuum%2Fcontinuum-feature-ai/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":31403960,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-04-04T10:20:44.708Z","status":"ssl_error","status_checked_at":"2026-04-04T10:20:06.846Z","response_time":60,"last_error":"SSL_read: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai","continuum","continuum-feature","cuda","llm","ml","mlops","pytourch","unsloth"],"created_at":"2026-04-04T15:42:31.824Z","updated_at":"2026-04-04T15:42:31.905Z","avatar_url":"https://github.com/projectcontinuum.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"\u003cdiv align=\"center\"\u003e\n  \u003ch1\u003eContinuum Feature AI\u003c/h1\u003e\n  \u003cstrong\u003eAI/ML nodes for fine-tuning LLMs inside your \u003ca href=\"https://github.com/projectcontinuum/Continuum\"\u003eProject Continuum\u003c/a\u003e workflows\u003c/strong\u003e\n\u003c/div\u003e\n\n\u003cdiv align=\"center\"\u003e\n  \u003cimg src=\"https://img.shields.io/badge/Kotlin-2.1.0-blue?logo=kotlin\u0026logoColor=white\" alt=\"Kotlin\"\u003e\n  \u003cimg src=\"https://img.shields.io/badge/Python-Unsloth-yellow?logo=python\u0026logoColor=white\" alt=\"Python\"\u003e\n  \u003cimg src=\"https://img.shields.io/badge/LoRA-Fine--Tuning-purple\" alt=\"LoRA\"\u003e\n  \u003cimg src=\"https://img.shields.io/badge/JDK-21-red\" alt=\"JDK 21\"\u003e\n\u003c/div\u003e\n\n---\n\n## 🌐 Part of Project Continuum\n\nThis is the **AI/ML feature repository** for [Project Continuum](https://github.com/projectcontinuum/Continuum) — a distributed, crash-proof workflow execution platform. It provides nodes for training and fine-tuning large language models directly inside your visual workflows.\n\n---\n\n## 🔥 What Is This\n\nA 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.\n\nShips as a Spring Boot worker with an auto-managed Python virtual environment for ML execution.\n\n---\n\n## 🧪 Included Nodes\n\n### LLM Trainer (Unsloth)\n\nFine-tune Large Language Models using LoRA with Unsloth acceleration.\n\n| | |\n|---|---|\n| **Input** | Parquet table with instruction + response columns |\n| **Output** | Model info — path to LoRA adapter weights, base model, training config |\n| **Category** | Machine Learning, LLM Training |\n\n**Supported Base Models:**\n\n| Provider | Models |\n|----------|--------|\n| **Unsloth** (fastest) | Phi-4, Phi-4-mini-instruct, Mistral 7B, Llama 3/3.1/3.2, Gemma 2, Qwen 2.5 |\n| **Microsoft** | Phi-2, Phi-3-mini-4k-instruct |\n| **Meta** | Llama-2-7b, Llama-2-7b-chat |\n| **Google** | Gemma 2B, Gemma 7B |\n| **Qwen** | Qwen2-7B, Qwen2-7B-Instruct |\n| **TII** | Falcon 7B, Falcon 7B-instruct |\n| **Custom** | Any HuggingFace causal language model |\n\n**Configurable Parameters:**\n\n| Group | Parameters |\n|-------|-----------|\n| **Model** | Base model (HuggingFace ID), HuggingFace token for gated models |\n| **Data** | Input column, output column, system prompt |\n| **Training** | Epochs, batch size, learning rate, max sequence length, warmup steps, weight decay, gradient accumulation |\n| **LoRA** | Rank (r), alpha, dropout |\n| **Advanced** | 4-bit quantization, random seed, save steps, logging steps, Parquet batch size |\n\n**Key Features:**\n- Unsloth acceleration on Linux + CUDA (2x faster, 60% less memory)\n- Falls back to standard HuggingFace transformers on other platforms\n- 4-bit quantization for reduced memory usage\n- Real-time training progress streaming via Kafka\n- Auto-managed Python virtual environment\n\n---\n\n## 🐍 Python Environment\n\nThe Unsloth node executes training via a Python virtual environment that is **automatically created at startup** if missing.\n\n| Setting | Default |\n|---------|---------|\n| `com.continuum.feature.ai.unsloth-trainer.venv-path` | `~/.continuum/unsloth-env` |\n| `com.continuum.feature.ai.unsloth-trainer.cache-storage-path` | `./.continuum-cache/workflow-data` |\n\n**Required Python packages** (auto-installed): pyarrow, pandas, datasets, torch, transformers, peft, trl, accelerate, hf_transfer, sentencepiece, protobuf, bitsandbytes, unsloth (Linux + CUDA only).\n\n---\n\n## 📦 Dependencies\n\nShared libraries from [Continuum](https://github.com/projectcontinuum/Continuum) via GitHub Packages:\n\n| Dependency | Purpose |\n|-----------|---------|\n| `continuum-commons:0.0.1` | Base node model, data types, Parquet/S3 utilities |\n| `continuum-worker-springboot-starter:0.0.1` | Worker framework — registers nodes with Temporal |\n\n---\n\n## 🚀 Quick Start\n\n### Prerequisites\n\n- **JDK 21** — [Eclipse Temurin](https://adoptium.net/) recommended\n- **Python 3.10+** — For the Unsloth training environment\n- **Docker \u0026 Docker Compose** — For local infrastructure\n- **GitHub PAT** with `read:packages` scope\n- **(Optional) CUDA GPU** — For Unsloth acceleration\n\nSet environment variables:\n\n```bash\nexport GITHUB_USERNAME=your-github-username\nexport GITHUB_TOKEN=ghp_your-personal-access-token\n```\n\n### Run\n\n```bash\n# Start infrastructure (Temporal, Kafka, MinIO, API Server, Message Bridge)\ncd docker \u0026\u0026 docker compose up -d\n\n# Build\n./gradlew build\n\n# Start the AI worker (auto-creates Python venv on first run)\n./gradlew :worker:bootRun\n```\n\n---\n\n## 📁 Project Structure\n\n```\ncontinuum-feature-ai/\n├── features/\n│   └── continuum-feature-unsloth/            # Unsloth LLM trainer node\n│       ├── build.gradle.kts                  # Depends on continuum-commons\n│       └── src/main/kotlin/.../\n│           ├── AutoConfigure.kt              # Spring auto-configuration\n│           ├── node/\n│           │   └── UnslothTrainerNodeModel.kt\n│           └── python/\n│               └── PythonEnvironmentManager.kt\n├── worker/                                   # Spring Boot worker application\n│   ├── build.gradle.kts                      # Depends on starter + unsloth feature\n│   └── src/main/\n│       ├── kotlin/.../App.kt\n│       └── resources/application.yaml\n├── docker/                                   # Local development infrastructure\n│   └── docker-compose.yml\n├── settings.gradle.kts\n├── gradle.properties\n└── README.md\n```\n\n---\n\n## 🗺️ Roadmap\n\n- [x] LLM fine-tuning with Unsloth + LoRA\n- [x] Auto-managed Python virtual environment\n- [x] 4-bit quantization support\n- [ ] Inference node — run inference against fine-tuned or base models\n- [ ] Model evaluation node — automated benchmarking\n- [ ] Multi-GPU training support\n- [ ] More model architectures (vision, embedding)\n\n---\n\n## 🔗 Related Repositories\n\n| Repository | Description |\n|-----------|-------------|\n| [Continuum](https://github.com/projectcontinuum/Continuum) | Core backend — API server, worker framework, shared libraries |\n| [continuum-workbench](https://github.com/projectcontinuum/continuum-workbench) | Browser IDE — Eclipse Theia + React Flow workflow editor |\n| [continuum-feature-base](https://github.com/projectcontinuum/continuum-feature-base) | Base analytics nodes — data transforms, REST, scripting, anomaly detection |\n| **continuum-feature-ai** (this repo) | AI/ML nodes — LLM fine-tuning with Unsloth + LoRA |\n| [continuum-feature-template](https://github.com/projectcontinuum/continuum-feature-template) | Template — scaffold your own custom worker with nodes |\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fprojectcontinuum%2Fcontinuum-feature-ai","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fprojectcontinuum%2Fcontinuum-feature-ai","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fprojectcontinuum%2Fcontinuum-feature-ai/lists"}