| { | |
| "architectures": [ | |
| "SortformerOffline" | |
| ], | |
| "ats_weight": 0.5, | |
| "dtype": "float32", | |
| "fc_encoder_config": { | |
| "activation_dropout": 0.1, | |
| "attention_bias": true, | |
| "attention_dropout": 0.1, | |
| "conv_kernel_size": 9, | |
| "dropout": 0.1, | |
| "dropout_positions": 0.0, | |
| "hidden_act": "silu", | |
| "hidden_size": 512, | |
| "initializer_range": 0.02, | |
| "intermediate_size": 2048, | |
| "layerdrop": 0.1, | |
| "max_position_embeddings": 5000, | |
| "model_type": "sortformer_fc_encoder", | |
| "num_attention_heads": 8, | |
| "num_hidden_layers": 18, | |
| "num_key_value_heads": 8, | |
| "num_mel_bins": 80, | |
| "scale_input": true, | |
| "subsampling_conv_channels": 256, | |
| "subsampling_conv_kernel_size": 3, | |
| "subsampling_conv_stride": 2, | |
| "subsampling_factor": 8 | |
| }, | |
| "initializer_range": 0.02, | |
| "model_type": "sortformer", | |
| "modules_config": { | |
| "causal_attn_rate": 0.5, | |
| "causal_attn_rc": 30, | |
| "chunk_left_context": 1, | |
| "chunk_len": 188, | |
| "chunk_right_context": 1, | |
| "dropout_rate": 0.5, | |
| "fc_d_model": 512, | |
| "fifo_len": 0, | |
| "max_index": 10000, | |
| "min_pos_scores_rate": 0.5, | |
| "model_type": "sortformer_modules", | |
| "num_speakers": 4, | |
| "pred_score_threshold": 1e-06, | |
| "scores_add_rnd": 2.0, | |
| "scores_boost_latest": 0.5, | |
| "sil_threshold": 0.1, | |
| "spkcache_len": 188, | |
| "spkcache_sil_frames_per_spk": 5, | |
| "spkcache_update_period": 188, | |
| "strong_boost_rate": 0.3, | |
| "subsampling_factor": 8, | |
| "tf_d_model": 192, | |
| "weak_boost_rate": 0.7 | |
| }, | |
| "num_speakers": 4, | |
| "pil_weight": 0.5, | |
| "tf_encoder_config": { | |
| "activation_dropout": 0.5, | |
| "activation_function": "relu", | |
| "attention_dropout": 0.5, | |
| "d_model": 192, | |
| "dropout": 0.5, | |
| "encoder_attention_heads": 8, | |
| "encoder_ffn_dim": 768, | |
| "encoder_layerdrop": 0.5, | |
| "encoder_layers": 18, | |
| "init_std": 0.02, | |
| "initializer_range": 0.02, | |
| "layer_norm_eps": 1e-05, | |
| "max_source_positions": 1500, | |
| "model_type": "sortformer_tf_encoder", | |
| "num_mel_bins": 80, | |
| "scale_embedding": false | |
| }, | |
| "transformers_version": "5.0.0.dev0" | |
| } | |