Module commands

Expand description

CLI Commands - Ollama-style interface

Modules§

bench: Bench command — single-process CLI benchmark (no HTTP).
bench_serve: Tokenizer-aware HTTP bench client — emits the canonical schema defined in crates/ferrum-bench-core and gated by docs/bench/PLAYBOOK.md § 7.
doctor: Read-only first-run diagnostics.
embed: Embed command - Generate embeddings using BERT models
list: List command - Show downloaded models
pull: Pull command - Download a model from HuggingFace Hub
replay_bundle: Offline request replay bundle validation.
run: Run command - Interactive chat with a model (ollama-style)
serve: Serve command - Start the HTTP inference server
stop: Stop command - Stop the running server
transcribe: Transcribe command - Speech-to-text using Whisper models
tts: TTS command - Text-to-speech using Qwen3-TTS models
vnext_checkpoint
vnext_determinism: CUDA vNext determinism evidence collection.