CLI and API Reference#
Reference pages for the FlashDreams command-line interface and Python APIs.
The unified flashdreams-run entry point: listing runner slugs,
inspecting a runner’s options, and launching single- and multi-GPU
inference.
The low-level kernels and process-group utilities that integrations share: attention, the block-structured KV cache, and distributed helpers.
The swappable abstractions every integration plugs into: the config system, the encoder / diffusion-model / decoder triple, and the streaming inference pipeline that drives them.
The two public layers a model integration is built from: pipelines that define model behavior and runners that define CLI-facing I/O.
The runner / pipeline building blocks for integration-driven serving, with LingBot-World as the canonical interactive-transport reference.