CLI Reference

The kt command is the user-facing CLI from kt-kernel.

Main Commands

Command	Role
`kt version`	Show version and environment information.
`kt doctor`	Diagnose environment issues.
`kt run`	Start a model inference server.
`kt chat`	Chat with a running model.
`kt model`	Manage model registry and model paths.
`kt config`	Manage CLI configuration.
`kt quant`	Quantize model weights where supported.
`kt bench` / `kt microbench`	Run benchmarks.
`kt sft`	Fine-tuning helper surface for LLaMA-Factory workflows.

kt version
kt doctor
kt model list
kt model search m2
kt run m2.1 --dry-run
kt run m2.1

Option	Role
`--host`, `--port`	Server bind address.
`--gpu-experts`	Registry-level alias for GPU expert placement.
`--cpu-threads`	Registry-level alias for CPU worker count.
`--tensor-parallel-size`	Tensor parallel size.
`--kt-method`	Override KT method when appropriate.
`--attention-backend`	Override attention backend.
`--max-total-tokens`	Token memory limit.
`--dry-run`	Print launch command without starting the server.

Use kt run --help in the target environment for the definitive command surface.