DeepForge

DeepForge is a monorepo of production-ready deep learning project templates. It includes:

• PyTorch Research: experiment scaffold with configs, training loop, tests.

• FastAPI Service: deploy trained models as APIs with Docker + CI/CD.

• Kaggle Baseline: clean starter kit for competitions with notebooks + pipelines.

All templates follow best practices: uv package manager, ruff, mypy, pytest, pre-commit, and GitHub Actions CI. Perfect for researchers, students, and engineers who want to go from prototype to production quickly and cleanly.

Tagline

• Production-ready deep learning templates for research, APIs, and Kaggle.

📁 Repository layout

DL-Project-Templates/ ├─ README.md ├─ pyproject.toml ├─ uv.lock ├─ Dockerfile ├─ docker-compose.yml ├─ Taskfile.yml ├─ main.py ├─ scripts/ │ ├─ train.py │ ├─ export.py │ └─ benchmark.py ├─ app/ │ ├─ init.py │ ├─ main.py │ ├─ api/ │ │ ├─ init.py │ │ ├─ routes_health.py │ │ ├─ routes_inference.py │ │ └─ routes_train.py │ ├─ core/ │ │ ├─ config.py │ │ └─ logger.py │ ├─ data/ │ │ └─ raw/.gitkeep │ ├─ models/ │ │ ├─ init.py │ │ ├─ model.py │ │ └─ weights/.gitkeep │ ├─ schemas/ │ │ ├─ init.py │ │ ├─ request.py │ │ └─ response.py │ ├─ services/ │ │ ├─ init.py │ │ ├─ loader.py │ │ ├─ inference_service.py │ │ └─ training_service.py │ └─ utils/ │ ├─ init.py │ ├─ file_utils.py │ ├─ preprocessing.py │ └─ postprocessing.py └─ tests/ ├─ init.py ├─ test_api.py ├─ test_end_to_end.py ├─ test_inference.py ├─ test_inference_sample.py └─ test_training.py

🔗 Root Level

• README.md → Overview and usage instructions for the microservice.

• pyproject.toml → Python packaging, dependencies, and dev tools config.

• uv.lock → Lockfile to pin exact versions of dependencies.

• Dockerfile → Container definition for running the API in production.

• docker-compose.yml → Local dev/CI orchestration (API, volumes, ports).

• Taskfile.yml → Task runner shortcuts (e.g. task run, task lint).

• main.py → Entry point for scripts (can load/run app or tasks).

• scripts/ → Standalone scripts for model lifecycle tasks (train, export, benchmark).

• tests/ → Unit + integration tests for endpoints, inference, and training.

🗂 app/

• init.py → Marks folder as a package.

• main.py → FastAPI application instance and router wiring.

• app/api/

• init.py → Package marker.

• routes_health.py → Health endpoint for monitoring (/api/health).

• routes_inference.py → Inference endpoint (/api/predict).

• routes_train.py → Training endpoint (/api/train).

• app/core/

• config.py → Centralized settings (paths, env vars, seeds).

• logger.py → Logger setup (Loguru) for consistent logging.

• app/data/

• raw/ → Placeholder for raw datasets used in training.

• .gitkeep → Ensures empty dir is tracked in git.

• app/models/

• init.py → Package marker.

• model.py → ML model definition (pipeline, architecture).

• weights/ → Directory for trained artifacts (.joblib, .pkl, .h5).

• app/schemas/

• init.py → Package marker.

• request.py → Pydantic schema for request validation (input features).

• response.py → Pydantic schema for response (predicted label, probabilities).

• app/services/

• init.py → Package marker.

• loader.py → Loads latest model artifact from weights/.

• inference_service.py → Contains inference logic (predict_proba).

• training_service.py → Training pipeline, saves new model artifacts.

• app/utils/

• init.py → Package marker.

• file_utils.py → Helpers for file/directory operations.

• preprocessing.py → Input preprocessing (clean/transform features).

• postprocessing.py → Output postprocessing (map probabilities to labels).

• ⚙️ scripts/

• train.py → CLI to trigger training and save a model artifact.

• export.py → CLI to export trained model to ONNX/TorchScript (future extension).

• benchmark.py → CLI to benchmark endpoint latency/throughput.

• 🧪 tests/

• init.py → Package marker.

• test_api.py → Tests API routes (health, predict).

• test_end_to_end.py → Tests full workflow (train → predict).

• test_inference.py → Unit tests for inference logic.

• test_inference_sample.py → Sample-based inference test with known inputs.

• test_training.py → Unit tests for training logic and artifact creation.

🧩 Template: templates/fastapi-ml-microservice

A complete FastAPI microservice for training + inference, mirroring the layout you shared. Includes health, training, and inference endpoints; model loader; preprocessing; and test suite.