ming-3.0-tts-tiny-glade / audio /metadata.json
coder543's picture
Publish Ming 3.0 TTS Tiny hybrid conversion for Glade
98f824c verified
Raw History Blame Contribute Delete
9.61 kB
{
"hidden_size": 896,
"layers": 24,
"attention_heads": 14,
"kv_heads": 2,
"head_dimension": 64,
"rope_theta": 1000000,
"rms_norm_epsilon": 1e-06,
"codec_cache_capacity": 64,
"codec_frame_capacity": 32,
"codec_attention_window": 64,
"codec_fft_size": 3528,
"codec_hop_length": 882,
"codec_padding": 1323,
"codec_upsample_factor": 4,
"codec_upsample_align_corners": false,
"codec_group_layers": 12,
"precision": "w8a16",
"normalization": "scaled-fp16",
"attention": "repeated-kv",
"projection_reduction_channels": 256,
"preferred_compute_unit_kind": "ane",
"schema_version": 1,
"model_family": "mingTinyTTSAudio",
"sample_rate": 44100,
"channels": 1,
"latent_dimension": 64,
"condition_dimension": 1536,
"flow_hidden_size": 1024,
"flow_layers": 16,
"aggregator_hidden_size": 1024,
"aggregator_layers": 8,
"flow_projection_reduction_channels": null,
"resources": "resources.json",
"resource_data": "resources.f32",
"codec_window_resource": "audio.decoder.head.istft.window",
"stop_projection": "stop_head",
"speaker_projection": "spk_head",
"time_features_resource": "flow.time_features",
"timesteps_resource": "flow.timesteps",
"assets": [
{
"name": "aggregator",
"functions": [
{
"name": "main",
"inputs": {
"latent": [
1,
64,
1,
1
]
},
"outputs": [
"hidden"
],
"output_shapes": {
"hidden": [
1,
1536,
1,
1
]
}
}
]
},
{
"name": "flow",
"functions": [
{
"name": "main",
"inputs": {
"noise": [
1,
64,
1,
1
],
"condition": [
1,
1536,
1,
1
],
"history": [
1,
64,
1,
1
],
"time_features": [
1,
1024,
1,
1
]
},
"outputs": [
"velocity"
],
"output_shapes": {
"velocity": [
2,
64,
1,
1
]
}
}
]
},
{
"name": "codec-0",
"functions": [
{
"name": "main",
"inputs": {
"hidden": [
1,
896,
1,
32
],
"cos": [
1,
64,
1,
32
],
"sin": [
1,
64,
1,
32
],
"mask": [
1,
96,
1,
32
],
"keys": [
1,
1536,
1,
64
],
"values": [
1,
1536,
1,
64
]
},
"outputs": [
"hidden",
"next_keys",
"next_values"
],
"output_shapes": {
"hidden": [
1,
896,
1,
32
],
"next_keys": [
1,
1536,
1,
32
],
"next_values": [
1,
1536,
1,
32
]
}
}
],
"family": "codec",
"layer_start": 0,
"layers": 12,
"cache_capacity": 64,
"window": 64
},
{
"name": "codec-1",
"functions": [
{
"name": "main",
"inputs": {
"hidden": [
1,
896,
1,
32
],
"cos": [
1,
64,
1,
32
],
"sin": [
1,
64,
1,
32
],
"mask": [
1,
96,
1,
32
],
"keys": [
1,
1536,
1,
64
],
"values": [
1,
1536,
1,
64
]
},
"outputs": [
"hidden",
"next_keys",
"next_values",
"spectrum"
],
"output_shapes": {
"hidden": [
1,
896,
1,
32
],
"next_keys": [
1,
1536,
1,
32
],
"next_values": [
1,
1536,
1,
32
],
"spectrum": [
1,
3530,
1,
32
]
}
}
],
"family": "codec",
"layer_start": 12,
"layers": 12,
"cache_capacity": 64,
"window": 64
},
{
"name": "encoder-0",
"functions": [
{
"name": "main",
"inputs": {
"hidden": [
1,
896,
1,
32
],
"cos": [
1,
64,
1,
32
],
"sin": [
1,
64,
1,
32
],
"mask": [
1,
96,
1,
32
],
"keys": [
1,
1536,
1,
64
],
"values": [
1,
1536,
1,
64
]
},
"outputs": [
"hidden",
"next_keys",
"next_values"
],
"output_shapes": {
"hidden": [
1,
896,
1,
32
],
"next_keys": [
1,
1536,
1,
32
],
"next_values": [
1,
1536,
1,
32
]
}
}
],
"family": "encoder",
"layer_start": 0,
"layers": 12,
"cache_capacity": 64,
"window": 64
},
{
"name": "encoder-1",
"functions": [
{
"name": "main",
"inputs": {
"hidden": [
1,
896,
1,
32
],
"cos": [
1,
64,
1,
32
],
"sin": [
1,
64,
1,
32
],
"mask": [
1,
96,
1,
32
],
"keys": [
1,
1536,
1,
64
],
"values": [
1,
1536,
1,
64
]
},
"outputs": [
"hidden",
"next_keys",
"next_values"
],
"output_shapes": {
"hidden": [
1,
896,
1,
32
],
"next_keys": [
1,
1536,
1,
32
],
"next_values": [
1,
1536,
1,
32
]
}
}
],
"family": "encoder",
"layer_start": 12,
"layers": 12,
"cache_capacity": 64,
"window": 64
},
{
"name": "encoder-aggregator-0",
"functions": [
{
"name": "main",
"inputs": {
"hidden": [
1,
896,
1,
40
],
"cos": [
1,
64,
1,
40
],
"sin": [
1,
64,
1,
40
],
"mask": [
1,
104,
1,
40
],
"keys": [
1,
512,
1,
64
],
"values": [
1,
512,
1,
64
]
},
"outputs": [
"hidden",
"next_keys",
"next_values"
],
"output_shapes": {
"hidden": [
1,
896,
1,
40
],
"next_keys": [
1,
512,
1,
40
],
"next_values": [
1,
512,
1,
40
]
}
}
],
"family": "encoder-aggregator",
"layer_start": 0,
"layers": 4,
"cache_capacity": 64,
"window": 64
}
]
}