preprocessor_config.json

{
  "config": {
    "audio_history_cfg": {
      "attn_qk_norm": true,
      "attn_scaled_cosine": false,
      "clip_attn_logit": null,
      "dropout_broadcast_dims": [
        -2
      ],
      "dropout_rate": 0.0,
      "droppath_rate": 0.0,
      "emb_dim": 768,
      "float32_attention_logits": true,
      "head_dim": 64,
      "latents_size": 16,
      "layer_drop": 0.0,
      "max_frames": 8,
      "mlp_activations": [
        "gelu"
      ],
      "mlp_dim": 2048,
      "num_heads": 12,
      "num_layers": 2,
      "resampler_type": "perceiver",
      "xattention_index": [
        0,
        1
      ],
      "xattn_qk_norm": true,
      "xattn_scaled_cosine": false
    },
    "audio_vit_cfg": {
      "default_input_size": [
        256,
        128
      ],
      "dropout_broadcast_dims": [],
      "dropout_rate": 0.0,
      "emb_dim": 768,
      "float32_attention_logits": true,
      "head_dim": 64,
      "mlp_activations": [
        "gelu"
      ],
      "mlp_dim": 3072,
      "num_heads": 12,
      "num_layers": 11,
      "patch_size": 16,
      "pos_patch_size": 16,
      "transpose_input": true,
      "vit_embed": true
    },
    "audio_vqgan": {
      "act_fn": "relu",
      "attention_dropout_rate": 0.0,
      "checkpoint_path": "",
      "decoder_head_dim": 64,
      "decoder_hidden_size": 512,
      "decoder_mlp_dim": 2048,
      "decoder_num_heads": 8,
      "decoder_num_layers": 8,
      "default_input_size": [
        128,
        256
      ],
      "dropout_rate": 0.0,
      "droppath_rate": 0.0,
      "encoder_head_dim": 64,
      "encoder_hidden_size": 512,
      "encoder_mlp_dim": 2048,
      "encoder_num_heads": 8,
      "encoder_num_layers": 8,
      "output_channel": 1,
      "patch_size": [
        8,
        8
      ],
      "proj_dim": 32,
      "use_bias": false,
      "use_decoder": true,
      "vocab_size": 8192
    },
    "freeze_vit": true,
    "image_history_cfg": {
      "attn_qk_norm": true,
      "attn_scaled_cosine": false,
      "clip_attn_logit": null,
      "dropout_broadcast_dims": [
        -2
      ],
      "dropout_rate": 0.0,
      "droppath_rate": 0.0,
      "emb_dim": 768,
      "float32_attention_logits": true,
      "head_dim": 64,
      "latents_size": 32,
      "layer_drop": 0.0,
      "max_frames": 8,
      "mlp_activations": [
        "gelu"
      ],
      "mlp_dim": 2048,
      "num_heads": 12,
      "num_layers": 2,
      "resampler_type": "perceiver",
      "xattention_index": [
        0,
        1
      ],
      "xattn_qk_norm": true,
      "xattn_scaled_cosine": false
    },
    "image_vit_cfg": {
      "default_input_size": [
        256,
        256
      ],
      "dropout_broadcast_dims": [],
      "dropout_rate": 0.0,
      "emb_dim": 768,
      "float32_attention_logits": true,
      "head_dim": 64,
      "mlp_activations": [
        "gelu"
      ],
      "mlp_dim": 3072,
      "num_heads": 12,
      "num_layers": 11,
      "num_pos": 197,
      "patch_size": 16,
      "pos_patch_size": 16
    },
    "image_vqgan": {
      "attn_resolutions": [
        32
      ],
      "ch": 128,
      "ch_mult": [
        1,
        2,
        2,
        4
      ],
      "checkpoint_path": "",
      "default_input_size": [
        256,
        256
      ],
      "double_z": false,
      "dropout": 0,
      "embed_dim": 4,
      "in_channels": 3,
      "n_embed": 16384,
      "num_res_blocks": 2,
      "out_ch": 3,
      "patch_size": [
        8,
        8
      ],
      "resolution": 256,
      "z_channels": 4
    },
    "input_modalities": [
      "text",
      "image",
      "image_history",
      "audio",
      "audio_history"
    ],
    "sequence_length": {
      "audio_history_input_samples": 128,
      "audio_input_samples": 128,
      "image_history_input_samples": 256,
      "image_input_samples": 576,
      "is_training": true,
      "num_frames": 4
    },
    "t5_config": {
      "audio_history_pos_emb": "llama_rope",
      "audio_patch_size": 16,
      "audio_pos_emb": "llama_rope",
      "audio_vit_patch_size": 16,
      "audio_vocab_size": 8320,
      "dalle_attn_mask": true,
      "decoder_max_audio_length": 512,
      "decoder_max_image_length": 1024,
      "decoder_max_text_length": 512,
      "decoder_xattention_internval": 1,
      "default_audio_history_vit_size": [
        256,
        128
      ],
      "default_audio_size": [
        256,
        128
      ],
      "default_audio_vit_size": [
        256,
        128
      ],
      "default_image_history_vit_size": [
        256,
        256
      ],
      "default_image_size": [
        256,
        256
      ],
      "default_image_vit_size": [
        384,
        384
      ],
      "dropout_broadcast_dims": [
        -2
      ],
      "dropout_rate": 0.0,
      "dynamic_unk_mask": true,
      "emb_dim": 1024,
      "encoder_max_audio_length": 128,
      "encoder_max_image_length": 576,
      "encoder_max_text_length": 512,
      "float32_attention_logits": true,
      "head_dim": 64,
      "image_history_pos_emb": "llama_rope",
      "image_patch_size": 16,
      "image_pos_emb": "llama_rope",
      "image_tokenizer_type": "vqgan",
      "image_vit_patch_size": 16,
      "image_vocab_size": 16512,
      "logits_via_embedding": true,
      "mlp_activations": [
        "silu",
        "linear"
      ],
      "mlp_dim": 2816,
      "num_decoder_layers": 24,
      "num_encoder_layers": 24,
      "num_heads": 16,
      "qk_norm": true,
      "text_pos_emb": "llama_rope",
      "vocab_size": 33280
    },
    "target_modalities": [
      "text",
      "image",
      "audio"
    ],
    "use_audio_history_vit": true,
    "use_audio_vit": true,
    "use_image_history_vit": true,
    "use_image_vit": true
  },
  "sequence_length": {
    "audio_history_input_samples": 128,
    "audio_input_samples": 128,
    "image_history_input_samples": 256,
    "image_input_samples": 576,
    "is_training": true,
    "num_frames": 4
  }
}