cavl_config.json

{
  "cut_layer": 27,
  "projection_output_dim": 1536,
  "max_num_image_tokens": 12,
  "hidden_dim": 1536,
  "loss_type": "subcenter_cosface",
  "head_type": "mlp",
  "pooler_type": "cross_modal",
  "num_queries": 1,
  "margin": 0.35,
  "scale": 32,
  "num_sub_centers": 3,
  "std": 0.05,
  "embedding_prompt": "<image> Analyze the provided document image and give me its visual description based on: Shapes and Elements: presence of graphical components, tables, sections, headers, and any other visual elements. Layout Consistency: Evaluate the spatial arrangement of text blocks, margins, and alignments. Content Type: Ensure the document types of content (e.g., tables, forms, paragraphs), regardless of specific wording."
}