granite.build

Use a new Granite.build build.yaml for any compute workload — model training (SFT/LoRA/EPT), inference/serving, data generation/processing, evaluation, or arbitrary compute. Targets, steps, inputs/outputs (artifacts), compute resources, and how they resolve within a space. Use when asked to create, write, or scaffold a new build, build.yaml, or pipeline.

Look up authoritative Granite.build documentation that ships inside the granite.build repo (build.yaml schema, steps, CLI, troubleshooting, glossary). Use when unsure about a Granite.build field, option, command, concept, or error, or when another gb skill says to consult the docs.

Clone, set up, and run the Granite.build standalone gbserver. Use when asked to set up, install, start, or run gbserver / the Granite.build standalone server and when running builds.