Upload config.json with huggingface_hub
Browse files- config.json +34 -28
config.json
CHANGED
|
@@ -6,45 +6,51 @@
|
|
| 6 |
"language": ["en"],
|
| 7 |
"architectures": ["VuiForConditionalGeneration"],
|
| 8 |
"model_files": {
|
| 9 |
-
"base": "vui-
|
| 10 |
-
"
|
| 11 |
-
"cohost": "vui-cohost-100m.pt",
|
| 12 |
-
"cohost_alt": "ckpts-vui-cohost-100m.pt",
|
| 13 |
-
"tokenizer": "fluac-22hz-22khz.pt"
|
| 14 |
},
|
| 15 |
"model_variants": {
|
| 16 |
-
"vui-
|
| 17 |
-
"description": "
|
| 18 |
-
"file": "vui-
|
| 19 |
-
"
|
| 20 |
-
|
| 21 |
-
"vui-abraham-100m": {
|
| 22 |
-
"description": "Single speaker model with context awareness",
|
| 23 |
-
"file": "vui-abraham-100m.pt",
|
| 24 |
-
"size_mb": 198
|
| 25 |
-
},
|
| 26 |
-
"vui-cohost-100m": {
|
| 27 |
-
"description": "Two speakers that can interact with each other",
|
| 28 |
-
"file": "vui-cohost-100m.pt",
|
| 29 |
-
"size_mb": 198
|
| 30 |
}
|
| 31 |
},
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
"tokenizer_config": {
|
| 33 |
-
"
|
| 34 |
-
"
|
| 35 |
-
"file": "fluac-22hz-22khz.pt",
|
| 36 |
-
"size_mb": 307
|
| 37 |
},
|
| 38 |
-
"
|
| 39 |
-
"
|
| 40 |
-
"
|
| 41 |
},
|
| 42 |
"capabilities": [
|
| 43 |
"text-to-speech",
|
| 44 |
-
"conversational-speech",
|
| 45 |
"voice-cloning",
|
|
|
|
| 46 |
"on-device-inference"
|
| 47 |
],
|
| 48 |
"torch_dtype": "float32",
|
| 49 |
"framework": "pytorch"
|
| 50 |
-
}
|
|
|
|
| 6 |
"language": ["en"],
|
| 7 |
"architectures": ["VuiForConditionalGeneration"],
|
| 8 |
"model_files": {
|
| 9 |
+
"base": "vui-nano.safetensors",
|
| 10 |
+
"tokenizer": "HuggingFaceTB/SmolLM2-135M"
|
|
|
|
|
|
|
|
|
|
| 11 |
},
|
| 12 |
"model_variants": {
|
| 13 |
+
"vui-nano": {
|
| 14 |
+
"description": "Streaming conversational TTS — Llama-style backbone with an RQ-Transformer acoustic head, paired with the Qwen3-TTS codec.",
|
| 15 |
+
"file": "vui-nano.safetensors",
|
| 16 |
+
"params": 305356033,
|
| 17 |
+
"size_mb": 1166
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
}
|
| 19 |
},
|
| 20 |
+
"model_config": {
|
| 21 |
+
"d_model": 768,
|
| 22 |
+
"n_layers": 22,
|
| 23 |
+
"n_heads": 8,
|
| 24 |
+
"n_kv_heads": null,
|
| 25 |
+
"bias": false,
|
| 26 |
+
"use_rotary_emb": true,
|
| 27 |
+
"rope_theta": 10000.0,
|
| 28 |
+
"use_rq_transformer": true,
|
| 29 |
+
"rq_n_layers": 5,
|
| 30 |
+
"rq_d_model": 768,
|
| 31 |
+
"rq_n_heads": 8,
|
| 32 |
+
"codebook_size": 2048,
|
| 33 |
+
"n_quantizers": 16,
|
| 34 |
+
"codec_hz": 12.5,
|
| 35 |
+
"spk_emb_dim": 1024,
|
| 36 |
+
"max_text_tokens": 100,
|
| 37 |
+
"max_audio_tokens": 100
|
| 38 |
+
},
|
| 39 |
"tokenizer_config": {
|
| 40 |
+
"base_tokenizer": "HuggingFaceTB/SmolLM2-135M",
|
| 41 |
+
"add_time_tokens": false
|
|
|
|
|
|
|
| 42 |
},
|
| 43 |
+
"audio_codec": {
|
| 44 |
+
"name": "qwen3-tts",
|
| 45 |
+
"native_sample_rate": 24000
|
| 46 |
},
|
| 47 |
"capabilities": [
|
| 48 |
"text-to-speech",
|
| 49 |
+
"conversational-speech",
|
| 50 |
"voice-cloning",
|
| 51 |
+
"streaming",
|
| 52 |
"on-device-inference"
|
| 53 |
],
|
| 54 |
"torch_dtype": "float32",
|
| 55 |
"framework": "pytorch"
|
| 56 |
+
}
|