| { |
| "model": { |
| "fm_decoder_downsampling_factor": [1, 2, 4, 2, 1], |
| "fm_decoder_num_layers": [1, 2, 3, 2, 1], |
| "fm_decoder_cnn_module_kernel": [31, 15, 7, 15, 31], |
| "fm_decoder_feedforward_dim": 1792, |
| "fm_decoder_feedforward_type": "depthwise_separable_conv", |
| "fm_decoder_feedforward_depthwise_kernel": 7, |
| "fm_decoder_num_heads": 4, |
| "fm_decoder_num_kv_heads": 2, |
| "fm_decoder_dim": 512, |
| "fm_decoder_use_nonlin_attention": false, |
| "fm_decoder_position_encoding_type": "rope", |
| "fm_decoder_rope_base": 10000.0, |
| "fm_decoder_convnext_num_layers": [1, 1, 2, 1, 1], |
| "fm_decoder_convnext_intermediate_dim": 1024, |
| "fm_decoder_convnext_kernel_size": 7, |
| "fm_decoder_convnext_dilation_pattern": [1, 2, 4, 8], |
| "fm_decoder_parameter_sharing_mode": "none", |
| "text_encoder_num_layers": 4, |
| "text_encoder_feedforward_dim": 512, |
| "text_encoder_feedforward_type": "standard", |
| "text_encoder_cnn_module_kernel": 9, |
| "text_encoder_num_heads": 4, |
| "text_encoder_dim": 256, |
| "text_refinement_num_layers": 3, |
| "text_refinement_intermediate_dim": 512, |
| "text_refinement_kernel_size": 7, |
| "text_refinement_dilation_pattern": [1, 2, 4], |
| "query_head_dim": 32, |
| "value_head_dim": 12, |
| "pos_head_dim": 4, |
| "pos_dim": 48, |
| "time_embed_dim": 256, |
| "text_embed_dim": 256, |
| "feat_dim": 100 |
| }, |
| "feature": { |
| "sampling_rate": 24000, |
| "type": "vocos" |
| } |
| } |
|
|
