Skip to content

CLI reference

Subcommand Purpose
mistralrs serve Start HTTP/MCP server and (optionally) the UI at /ui
mistralrs run Run model in interactive mode, or one-shot mode with -i
mistralrs completions Generate shell completions
mistralrs quantize Generate UQFF quantized model file
mistralrs uqff Inspect, report, or verify UQFF artifacts
mistralrs doctor Run system diagnostics and environment checks
mistralrs tune Recommend quantization + device mapping for a model. Rejects --quant auto; pass --quant <level> or --isq <level> to bias the recommendation toward a specific quantization target. Adapter options are rejected because adapter memory is not included in the estimate
mistralrs login Authenticate with Hugging Face Hub
mistralrs cache Manage the Hugging Face model cache
mistralrs bench Run performance benchmarks for base or LoRA model generation
mistralrs from-config Run from a full TOML configuration file
mistralrs update Update or migrate an install using the installer
mistralrs uninstall Remove an installer-managed install
Option Default Description
--seed <SEED> Random seed for reproducibility
-l, --log <LOG> Log all requests and responses to this file
--token-source <TOKEN_SOURCE> cache Token source for Hugging Face authentication. Formats: literal:<token>, env:<var>, path:<file>, cache, none
-v, --verbose 0 Increase logging verbosity. Use -v for debug and -vv for trace-level internals