# hipfire

hipfire is an AMD GPU-native inference engine, written in Rust, for running large language models locally with an Ollama-style command-line interface and an OpenAI-compatible API.

hipfire runs LLM inference on AMD GPUs using the RDNA architecture family, RDNA1 through RDNA4, and also supports CDNA. It provides an Ollama-style workflow, with pull, serve and chat commands for named models, and exposes an OpenAI-compatible API from its local daemon.

A component called Redline is hipfire's own dispatch and replay layer: it records the kernel graph a model run produces, tracks resource dependencies, and replays validated paths directly through ROCr queues; anything it cannot validate falls back to ordinary HIP dispatch. Curated model weights, including a performance-oriented Qwen build, are published through the project's own Hugging Face collection and a per-model registry.

The current release also adds image generation, using the FLUX model family, alongside text inference.

Website: https://github.com/warpfront/hipfire