Open-source projects, inference systems, research tooling, and datasets from the Rootcomputer team. For everything else, visit our GitHub.
The 50,000-token Byte Pair Encoding (BPE) vocabulary used across the Haiku Mini model line.
A minimal Python inference server demonstrating the runtime structure used by Rootcomputer’s Haiku models.
Curated conversational dataset snippet pulled from the Haiku Mini training dataset.