Unsloth

Turn an Unsloth request into a reproducible installation, bounded experiment, or verified model artifact. Unsloth provides an optimized Python training stack (Core), a browser interface (Studio), and a native application (Desktop). Capabilities differ by product, backend, model, and release. Read the relevant official page before selecting a new hardware or model path; never promote a vendor speed or minimum-memory claim into a promise about the user's workload.

When not to use

• For deciding whether ML, retrieval, or fine-tuning is appropriate, use ml-engineering; return here for Unsloth execution.
• For operating exported models in a standalone engine, use llama-cpp or vllm. This skill owns the export and parity checks at that boundary.
• For container platform design use platform-engineering; for Docker Compose mechanics use docker-compose. Do not turn this skill into a platform manual.

Start here