Developer Guides#
The end-to-end computation flow: warmup, CUDA-graph capture, the autoregressive-step body, the ring-attention shard group, and finalize. The mental model the rest of the project assumes.
How every overridable field is surfaced as a CLI flag, how method defaults compose, and how to layer overrides on top.
How public runner names are registered, parsed, matched to manifests, and dispatched to integration-owned demo launch modes.
The entry-point surface a new method ships against: what to subclass, what to register, and where the parity tests live.
How to run command-backed local benchmarks that capture logs, MP4s, metrics, environment metadata, and an HTML report.
The low-level quantization and optimized multi-head attention building blocks used to accelerate streaming video models.
Where these guides fit#
These guides are conceptual. For a specific method, see its per-model page under Models; for the per-symbol reference, see CLI and API Reference; and for the two-command path from install to a generated clip, see Get Started.