| { |
| "architectures": [ |
| "ClapModel" |
| ], |
| "audio_config": { |
| "aff_block_r": 4, |
| "attention_dropout": 0.1, |
| "attention_probs_dropout_prob": 0.0, |
| "depths": [ |
| 2, |
| 2 |
| ], |
| "drop_path_rate": 0.0, |
| "dropout": 0.1, |
| "enable_fusion": false, |
| "enable_patch_layer_norm": true, |
| "flatten_patch_embeds": true, |
| "freq_ratio": 2, |
| "fusion_type": null, |
| "hidden_act": "gelu", |
| "hidden_dropout_prob": 0.1, |
| "hidden_size": 32, |
| "image_size": 60, |
| "initializer_factor": 1.0, |
| "initializer_range": 0.02, |
| "intermediate_size": 37, |
| "layer_norm_eps": 1e-05, |
| "mlp_ratio": 4.0, |
| "model_type": "clap_audio_model", |
| "num_attention_heads": [ |
| 2, |
| 2 |
| ], |
| "num_channels": 3, |
| "num_classes": 527, |
| "num_hidden_layers": 2, |
| "num_mel_bins": 16, |
| "patch_embed_input_channels": 1, |
| "patch_embeds_hidden_size": 16, |
| "patch_size": 2, |
| "patch_stride": 2, |
| "projection_dim": 64, |
| "projection_hidden_act": "relu", |
| "qkv_bias": true, |
| "spec_size": 64, |
| "window_size": 4 |
| }, |
| "dtype": "float32", |
| "hidden_size": 32, |
| "initializer_factor": 1.0, |
| "logit_scale_init_value": 14.285714285714285, |
| "model_type": "clap", |
| "num_hidden_layers": 4, |
| "projection_dim": 64, |
| "projection_hidden_act": "relu", |
| "text_config": { |
| "attention_dropout": 0.1, |
| "attention_probs_dropout_prob": 0.1, |
| "bos_token_id": 0, |
| "dropout": 0.1, |
| "eos_token_id": 2, |
| "hidden_act": "gelu", |
| "hidden_dropout_prob": 0.1, |
| "hidden_size": 32, |
| "initializer_factor": 1.0, |
| "initializer_range": 0.02, |
| "intermediate_size": 37, |
| "layer_norm_eps": 1e-12, |
| "max_position_embeddings": 512, |
| "model_type": "clap_text_model", |
| "num_attention_heads": 4, |
| "num_hidden_layers": 2, |
| "pad_token_id": 1, |
| "projection_dim": 64, |
| "projection_hidden_act": "relu", |
| "type_vocab_size": 1, |
| "vocab_size": 1024 |
| }, |
| "transformers_version": "5.15.0.dev0" |
| } |
|
|