Cog

Cog is an open-source tool that simplifies packaging machine learning models into standard, production-ready Docker containers. Instead of writing complex Dockerfiles or dealing with CUDA and Python dependency hell, you define your environment in a simple `cog.yaml` config and implement a Python class with a `run()` method. Cog automatically selects compatible NVIDIA base images, resolves CUDA/cuDNN/Python combos, and efficiently caches dependencies. It generates an OpenAPI schema from your model's Python type hints and spins up a high-performance Rust/Axum HTTP inference server. You can run models locally, build Docker images for deployment, serve via HTTP, or execute training scripts with `cog exec`. It also integrates with Replicate for cloud deployment. Built by the creators of Docker Compose, Cog is actively maintained with a growing community on Discord. It's ideal for ML engineers and researchers who want to ship models without wrestling with Dockerfiles, Flask, or CUDA configurations. Unlike generic Docker setups, Cog enforces best practices out of the box and provides a consistent interface from development to production.