| { | |
| "schema_version": 3, | |
| "language": "onnx_french", | |
| "bundle_name": "onnx_french", | |
| "sample_rate": 24000, | |
| "frame_rate": 12.5, | |
| "samples_per_frame": 1920, | |
| "latent_dim": 32, | |
| "conditioning_dim": 1024, | |
| "insert_bos_before_voice": true, | |
| "bos_before_voice_file": "bos_before_voice.npy", | |
| "max_token_per_chunk": 50, | |
| "remove_semicolons": true, | |
| "replace_characters": { | |
| "\"": "", | |
| "\u201c": "", | |
| "\u201d": "", | |
| "\u201e": "", | |
| "\u00ab": "", | |
| "\u00bb": "", | |
| "\u2019": "'", | |
| "\u2018": "'", | |
| ":": ",", | |
| "(": "", | |
| ")": "", | |
| "[": "", | |
| "]": "" | |
| }, | |
| "capitalize_first_letter": true, | |
| "append_terminal_punctuation": true, | |
| "default_temperature": 0.3, | |
| "pad_with_spaces_for_short_inputs": false, | |
| "model_recommended_frames_after_eos": null, | |
| "predefined_voices": [], | |
| "tokenizer_file": "tokenizer.model", | |
| "mimi_increment_per_frame": 16, | |
| "flow_lm_state_manifest": [ | |
| { | |
| "module": "transformer.layers.0.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 1000, | |
| 16, | |
| 64 | |
| ], | |
| "index": 0, | |
| "input_name": "state_0", | |
| "output_name": "out_state_0" | |
| }, | |
| { | |
| "module": "transformer.layers.0.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 1, | |
| "input_name": "state_1", | |
| "output_name": "out_state_1" | |
| }, | |
| { | |
| "module": "transformer.layers.1.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 1000, | |
| 16, | |
| 64 | |
| ], | |
| "index": 2, | |
| "input_name": "state_2", | |
| "output_name": "out_state_2" | |
| }, | |
| { | |
| "module": "transformer.layers.1.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 3, | |
| "input_name": "state_3", | |
| "output_name": "out_state_3" | |
| }, | |
| { | |
| "module": "transformer.layers.2.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 1000, | |
| 16, | |
| 64 | |
| ], | |
| "index": 4, | |
| "input_name": "state_4", | |
| "output_name": "out_state_4" | |
| }, | |
| { | |
| "module": "transformer.layers.2.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 5, | |
| "input_name": "state_5", | |
| "output_name": "out_state_5" | |
| }, | |
| { | |
| "module": "transformer.layers.3.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 1000, | |
| 16, | |
| 64 | |
| ], | |
| "index": 6, | |
| "input_name": "state_6", | |
| "output_name": "out_state_6" | |
| }, | |
| { | |
| "module": "transformer.layers.3.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 7, | |
| "input_name": "state_7", | |
| "output_name": "out_state_7" | |
| }, | |
| { | |
| "module": "transformer.layers.4.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 1000, | |
| 16, | |
| 64 | |
| ], | |
| "index": 8, | |
| "input_name": "state_8", | |
| "output_name": "out_state_8" | |
| }, | |
| { | |
| "module": "transformer.layers.4.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 9, | |
| "input_name": "state_9", | |
| "output_name": "out_state_9" | |
| }, | |
| { | |
| "module": "transformer.layers.5.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 1000, | |
| 16, | |
| 64 | |
| ], | |
| "index": 10, | |
| "input_name": "state_10", | |
| "output_name": "out_state_10" | |
| }, | |
| { | |
| "module": "transformer.layers.5.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 11, | |
| "input_name": "state_11", | |
| "output_name": "out_state_11" | |
| } | |
| ], | |
| "mimi_state_manifest": [ | |
| { | |
| "module": "decoder.model.0", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 0, | |
| "input_name": "state_0", | |
| "output_name": "out_state_0" | |
| }, | |
| { | |
| "module": "decoder.model.0", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 512, | |
| 6 | |
| ], | |
| "index": 1, | |
| "input_name": "state_1", | |
| "output_name": "out_state_1" | |
| }, | |
| { | |
| "module": "decoder.model.11", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 2, | |
| "input_name": "state_2", | |
| "output_name": "out_state_2" | |
| }, | |
| { | |
| "module": "decoder.model.11", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 64, | |
| 2 | |
| ], | |
| "index": 3, | |
| "input_name": "state_3", | |
| "output_name": "out_state_3" | |
| }, | |
| { | |
| "module": "decoder.model.2", | |
| "key": "partial", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 256, | |
| 6 | |
| ], | |
| "index": 4, | |
| "input_name": "state_4", | |
| "output_name": "out_state_4" | |
| }, | |
| { | |
| "module": "decoder.model.3.block.1", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 5, | |
| "input_name": "state_5", | |
| "output_name": "out_state_5" | |
| }, | |
| { | |
| "module": "decoder.model.3.block.1", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 256, | |
| 2 | |
| ], | |
| "index": 6, | |
| "input_name": "state_6", | |
| "output_name": "out_state_6" | |
| }, | |
| { | |
| "module": "decoder.model.3.block.3", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 7, | |
| "input_name": "state_7", | |
| "output_name": "out_state_7" | |
| }, | |
| { | |
| "module": "decoder.model.3.block.3", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 128, | |
| 0 | |
| ], | |
| "index": 8, | |
| "input_name": "state_8", | |
| "output_name": "out_state_8" | |
| }, | |
| { | |
| "module": "decoder.model.5", | |
| "key": "partial", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 128, | |
| 5 | |
| ], | |
| "index": 9, | |
| "input_name": "state_9", | |
| "output_name": "out_state_9" | |
| }, | |
| { | |
| "module": "decoder.model.6.block.1", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 10, | |
| "input_name": "state_10", | |
| "output_name": "out_state_10" | |
| }, | |
| { | |
| "module": "decoder.model.6.block.1", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 128, | |
| 2 | |
| ], | |
| "index": 11, | |
| "input_name": "state_11", | |
| "output_name": "out_state_11" | |
| }, | |
| { | |
| "module": "decoder.model.6.block.3", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 12, | |
| "input_name": "state_12", | |
| "output_name": "out_state_12" | |
| }, | |
| { | |
| "module": "decoder.model.6.block.3", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 64, | |
| 0 | |
| ], | |
| "index": 13, | |
| "input_name": "state_13", | |
| "output_name": "out_state_13" | |
| }, | |
| { | |
| "module": "decoder.model.8", | |
| "key": "partial", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 64, | |
| 4 | |
| ], | |
| "index": 14, | |
| "input_name": "state_14", | |
| "output_name": "out_state_14" | |
| }, | |
| { | |
| "module": "decoder.model.9.block.1", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 15, | |
| "input_name": "state_15", | |
| "output_name": "out_state_15" | |
| }, | |
| { | |
| "module": "decoder.model.9.block.1", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 64, | |
| 2 | |
| ], | |
| "index": 16, | |
| "input_name": "state_16", | |
| "output_name": "out_state_16" | |
| }, | |
| { | |
| "module": "decoder.model.9.block.3", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 17, | |
| "input_name": "state_17", | |
| "output_name": "out_state_17" | |
| }, | |
| { | |
| "module": "decoder.model.9.block.3", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 32, | |
| 0 | |
| ], | |
| "index": 18, | |
| "input_name": "state_18", | |
| "output_name": "out_state_18" | |
| }, | |
| { | |
| "module": "decoder_transformer.transformer.layers.0.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 8192, | |
| 8, | |
| 64 | |
| ], | |
| "index": 19, | |
| "input_name": "state_19", | |
| "output_name": "out_state_19" | |
| }, | |
| { | |
| "module": "decoder_transformer.transformer.layers.0.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 20, | |
| "input_name": "state_20", | |
| "output_name": "out_state_20" | |
| }, | |
| { | |
| "module": "decoder_transformer.transformer.layers.1.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 8192, | |
| 8, | |
| 64 | |
| ], | |
| "index": 21, | |
| "input_name": "state_21", | |
| "output_name": "out_state_21" | |
| }, | |
| { | |
| "module": "decoder_transformer.transformer.layers.1.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 22, | |
| "input_name": "state_22", | |
| "output_name": "out_state_22" | |
| }, | |
| { | |
| "module": "downsample.conv", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 23, | |
| "input_name": "state_23", | |
| "output_name": "out_state_23" | |
| }, | |
| { | |
| "module": "downsample.conv", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 512, | |
| 16 | |
| ], | |
| "index": 24, | |
| "input_name": "state_24", | |
| "output_name": "out_state_24" | |
| }, | |
| { | |
| "module": "encoder.model.0", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 25, | |
| "input_name": "state_25", | |
| "output_name": "out_state_25" | |
| }, | |
| { | |
| "module": "encoder.model.0", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 1, | |
| 6 | |
| ], | |
| "index": 26, | |
| "input_name": "state_26", | |
| "output_name": "out_state_26" | |
| }, | |
| { | |
| "module": "encoder.model.1.block.1", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 27, | |
| "input_name": "state_27", | |
| "output_name": "out_state_27" | |
| }, | |
| { | |
| "module": "encoder.model.1.block.1", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 64, | |
| 2 | |
| ], | |
| "index": 28, | |
| "input_name": "state_28", | |
| "output_name": "out_state_28" | |
| }, | |
| { | |
| "module": "encoder.model.1.block.3", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 29, | |
| "input_name": "state_29", | |
| "output_name": "out_state_29" | |
| }, | |
| { | |
| "module": "encoder.model.1.block.3", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 32, | |
| 0 | |
| ], | |
| "index": 30, | |
| "input_name": "state_30", | |
| "output_name": "out_state_30" | |
| }, | |
| { | |
| "module": "encoder.model.11", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 31, | |
| "input_name": "state_31", | |
| "output_name": "out_state_31" | |
| }, | |
| { | |
| "module": "encoder.model.11", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 512, | |
| 2 | |
| ], | |
| "index": 32, | |
| "input_name": "state_32", | |
| "output_name": "out_state_32" | |
| }, | |
| { | |
| "module": "encoder.model.3", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 33, | |
| "input_name": "state_33", | |
| "output_name": "out_state_33" | |
| }, | |
| { | |
| "module": "encoder.model.3", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 64, | |
| 4 | |
| ], | |
| "index": 34, | |
| "input_name": "state_34", | |
| "output_name": "out_state_34" | |
| }, | |
| { | |
| "module": "encoder.model.4.block.1", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 35, | |
| "input_name": "state_35", | |
| "output_name": "out_state_35" | |
| }, | |
| { | |
| "module": "encoder.model.4.block.1", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 128, | |
| 2 | |
| ], | |
| "index": 36, | |
| "input_name": "state_36", | |
| "output_name": "out_state_36" | |
| }, | |
| { | |
| "module": "encoder.model.4.block.3", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 37, | |
| "input_name": "state_37", | |
| "output_name": "out_state_37" | |
| }, | |
| { | |
| "module": "encoder.model.4.block.3", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 64, | |
| 0 | |
| ], | |
| "index": 38, | |
| "input_name": "state_38", | |
| "output_name": "out_state_38" | |
| }, | |
| { | |
| "module": "encoder.model.6", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 39, | |
| "input_name": "state_39", | |
| "output_name": "out_state_39" | |
| }, | |
| { | |
| "module": "encoder.model.6", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 128, | |
| 5 | |
| ], | |
| "index": 40, | |
| "input_name": "state_40", | |
| "output_name": "out_state_40" | |
| }, | |
| { | |
| "module": "encoder.model.7.block.1", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 41, | |
| "input_name": "state_41", | |
| "output_name": "out_state_41" | |
| }, | |
| { | |
| "module": "encoder.model.7.block.1", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 256, | |
| 2 | |
| ], | |
| "index": 42, | |
| "input_name": "state_42", | |
| "output_name": "out_state_42" | |
| }, | |
| { | |
| "module": "encoder.model.7.block.3", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 43, | |
| "input_name": "state_43", | |
| "output_name": "out_state_43" | |
| }, | |
| { | |
| "module": "encoder.model.7.block.3", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 128, | |
| 0 | |
| ], | |
| "index": 44, | |
| "input_name": "state_44", | |
| "output_name": "out_state_44" | |
| }, | |
| { | |
| "module": "encoder.model.9", | |
| "key": "first", | |
| "dtype": "bool", | |
| "fill": "ones", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 45, | |
| "input_name": "state_45", | |
| "output_name": "out_state_45" | |
| }, | |
| { | |
| "module": "encoder.model.9", | |
| "key": "previous", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 256, | |
| 6 | |
| ], | |
| "index": 46, | |
| "input_name": "state_46", | |
| "output_name": "out_state_46" | |
| }, | |
| { | |
| "module": "encoder_transformer.transformer.layers.0.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 8192, | |
| 8, | |
| 64 | |
| ], | |
| "index": 47, | |
| "input_name": "state_47", | |
| "output_name": "out_state_47" | |
| }, | |
| { | |
| "module": "encoder_transformer.transformer.layers.0.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 48, | |
| "input_name": "state_48", | |
| "output_name": "out_state_48" | |
| }, | |
| { | |
| "module": "encoder_transformer.transformer.layers.1.self_attn", | |
| "key": "cache", | |
| "dtype": "float32", | |
| "fill": "nan", | |
| "shape": [ | |
| 2, | |
| 1, | |
| 8192, | |
| 8, | |
| 64 | |
| ], | |
| "index": 49, | |
| "input_name": "state_49", | |
| "output_name": "out_state_49" | |
| }, | |
| { | |
| "module": "encoder_transformer.transformer.layers.1.self_attn", | |
| "key": "offset", | |
| "dtype": "int64", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1 | |
| ], | |
| "index": 50, | |
| "input_name": "state_50", | |
| "output_name": "out_state_50" | |
| }, | |
| { | |
| "module": "upsample.convtr", | |
| "key": "partial", | |
| "dtype": "float32", | |
| "fill": "zeros", | |
| "shape": [ | |
| 1, | |
| 512, | |
| 16 | |
| ], | |
| "index": 51, | |
| "input_name": "state_51", | |
| "output_name": "out_state_51" | |
| } | |
| ] | |
| } |