izwi pull - Izwi

Download a model from Hugging Face.

Synopsis

izwi pull <MODEL> [OPTIONS]

Downloads a model from the Hugging Face Hub and caches it locally. Supports resuming interrupted downloads.

Argument	Description
`<MODEL>`	Model variant to download

Option	Description	Default
`-f, --force`	Force re-download even if exists	—
`-y, --yes`	Download without confirmation	—

izwi pull Qwen3-TTS-12Hz-0.6B-Base

izwi pull Qwen3-TTS-12Hz-0.6B-Base --yes

izwi pull Qwen3-TTS-12Hz-0.6B-Base --force

Run izwi list to see all available models. Common models:

Model	Type	Size
`Qwen3-TTS-12Hz-0.6B-Base`	TTS (base)	~2.3 GB
`Qwen3-TTS-12Hz-0.6B-CustomVoice-4bit`	TTS (built-in voices)	~1.6 GB
`Qwen3-TTS-12Hz-1.7B-VoiceDesign-4bit`	TTS (voice design)	~2.2 GB
`Voxtral-4B-TTS-2603`	TTS (20 preset voices, CC BY-NC 4.0)	~8.1 GB
`Parakeet-TDT-0.6B-v3`	ASR (default)	~9.4 GB
`Qwen3-ASR-0.6B-GGUF`	ASR (compact)	~1.0 GB
`Nemotron-3.5-ASR-Streaming-0.6B`	ASR (NVIDIA multilingual `.nemo`)	~2.37 GB
`Granite-Speech-4.1-2B-Plus`	ASR (IBM rich transcription)	~4.2 GB
`Voxtral-Mini-4B-Realtime-2602`	ASR (offline transcription; realtime planned)	~8 GB
`Qwen3-8B-GGUF`	Chat	~5.2 GB
`Qwen3.5-4B`	Chat	~3.4 GB
`LFM2.5-1.2B-Instruct-GGUF`	Chat	~0.7 GB
`Gemma-3-1b-it`	Chat	~2.2 GB

If a download is interrupted, run the same command again. The download will resume from where it left off.