config.json

{
    "vit_hidden_dim": 768,
    "vit_inter_dim": 3072,
    "vit_patch_size": 16,
    "vit_img_size": 224,
    "vit_n_heads": 12,
    "vit_dropout": 0.0,
    "vit_n_blocks": 12,
    "vit_ln_eps": 1e-06,
    "vit_cls_flag": false,
    "vit_model_type": "google/siglip-base-patch16-224",
    "vae_enable": true,
    "vae_attn_resolutions": [
        3
    ],
    "vae_ch": 256,
    "vae_ch_mult": [
        1,
        2,
        2,
        4
    ],
    "vae_codebook_size": 32768,
    "vae_double_z": false,
    "vae_dropout": 0.0,
    "vae_embed_dim": 4,
    "vae_in_channels": 3,
    "vae_num_res_blocks": 2,
    "vae_out_channels": 3,
    "vae_temporal_downsample_factor": 4,
    "vae_torch_dtype": "float32",
    "vae_z_channels": 4,
    "vae_model_type": "BAAI/Emu3-VisionTokenizer",
    "lm_hidden_dim": 576,
    "lm_inter_dim": 1536,
    "lm_rms_eps": 1e-05,
    "lm_re_base": 100000,
    "lm_max_position_embeddings": 8192,
    "lm_vocab_size": 81922,
    "lm_n_heads": 9,
    "lm_n_kv_heads": 3,
    "lm_dropout": 0.0,
    "lm_n_blocks": 30,
    "lm_attn_scaling": 1.0,
    "TOTAL_SEQUENCE_LENGTH": 1280,
    "lm_max_length": 1280,
    "lm_use_tokens": true,
    "lm_tie_weights": true,
    "lm_model_type": "HuggingFaceTB/SmolLM2-135M",
    "lm_tokenizer": "HuggingFaceTB/cosmo2-tokenizer",
    "lm_eos_token_id": 0,
    "mp_pixel_shuffle_factor": 2,
    "vlm_load_backbone_weights": true,
    "vlm_checkpoint_path": "checkpoints/",
    "hf_repo_name": "blanchon/nanoVLM-222M"
}