| { | |
| "activation_dropout": 0, | |
| "activation_function": "gelu", | |
| "architectures": [ | |
| "Qwen3OmniMoeAudioEncoder" | |
| ], | |
| "attention_dropout": 0, | |
| "conv_chunksize": 500, | |
| "d_model": 1280, | |
| "downsample_hidden_size": 480, | |
| "dropout": 0, | |
| "dtype": "bfloat16", | |
| "encoder_attention_heads": 20, | |
| "encoder_ffn_dim": 5120, | |
| "encoder_layers": 32, | |
| "initializer_range": 0.02, | |
| "max_source_positions": 1500, | |
| "model_type": "qwen3_omni_moe_audio_encoder", | |
| "n_window": 50, | |
| "n_window_infer": 800, | |
| "num_hidden_layers": 32, | |
| "num_mel_bins": 128, | |
| "output_dim": 2048, | |
| "scale_embedding": false, | |
| "tf_legacy_loss": false, | |
| "transformers_version": "4.57.0.dev0", | |
| "use_bfloat16": false | |
| } | |