| { | |
| "model_type": "vui", | |
| "library_name": "vui", | |
| "pipeline_tag": "text-to-speech", | |
| "license": "apache-2.0", | |
| "language": [ | |
| "en" | |
| ], | |
| "architectures": [ | |
| "VuiForConditionalGeneration" | |
| ], | |
| "model_files": { | |
| "base": "vui-nano.safetensors", | |
| "tokenizer": "HuggingFaceTB/SmolLM2-135M" | |
| }, | |
| "model_variants": { | |
| "vui-nano": { | |
| "description": "Small, context-aware streaming TTS trained on real conversations — Llama-style backbone with an RQ-Transformer acoustic head over the Qwen3-TTS codec. Conditions on dialogue history (text and prior-turn audio) across a ~6-minute context.", | |
| "file": "vui-nano.safetensors", | |
| "params": 305356033, | |
| "size_mb": 611, | |
| "active_params": 218623489 | |
| } | |
| }, | |
| "model_config": { | |
| "d_model": 768, | |
| "n_layers": 22, | |
| "n_heads": 8, | |
| "n_kv_heads": null, | |
| "bias": false, | |
| "use_rotary_emb": true, | |
| "rope_theta": 10000.0, | |
| "use_rq_transformer": true, | |
| "rq_n_layers": 5, | |
| "rq_d_model": 768, | |
| "rq_n_heads": 8, | |
| "codebook_size": 2048, | |
| "n_quantizers": 16, | |
| "codec_hz": 12.5, | |
| "spk_emb_dim": 1024, | |
| "max_text_tokens": 100, | |
| "max_audio_tokens": 100 | |
| }, | |
| "tokenizer_config": { | |
| "base_tokenizer": "HuggingFaceTB/SmolLM2-135M", | |
| "add_time_tokens": false | |
| }, | |
| "audio_codec": { | |
| "name": "qwen3-tts", | |
| "native_sample_rate": 24000 | |
| }, | |
| "capabilities": [ | |
| "text-to-speech", | |
| "conversational-speech", | |
| "context-aware", | |
| "voice-cloning", | |
| "streaming", | |
| "on-device-inference", | |
| "cpu-inference" | |
| ], | |
| "torch_dtype": "bfloat16", | |
| "framework": "pytorch", | |
| "sample_rate": 24000 | |
| } | |