model_config.json

{
  "architecture": "bbox_indicator_visual_baseline",
  "decoder_memory": "[bbox_indicator_token, siglip_patch_tokens]",
  "run_mode": "full",
  "vision_model_name": "google/siglip2-large-patch16-256",
  "text_model_name": "facebook/mbart-large-50",
  "target_lang": "vi",
  "tgt_lang_code": "vi_VN",
  "bbox_indicator_ref": 1000.0,
  "bbox_indicator_dim": 4,
  "max_target_length": 64,
  "max_new_tokens": 30,
  "num_beams": 4,
  "repetition_penalty": 1.1,
  "no_repeat_ngram_size": 3,
  "length_penalty": 1.0,
  "train_batch_size": 64,
  "eval_batch_size": 64,
  "grad_accum_steps": 1,
  "effective_batch_size": 64,
  "num_epochs": 30,
  "learning_rate": 3e-05,
  "warmup_steps": 350,
  "max_grad_norm": 1.0,
  "weight_decay": 0.01,
  "label_smoothing": 0.1,
  "freeze_vision": true,
  "use_lora": true,
  "lora_r": 16,
  "lora_alpha": 32,
  "lora_dropout": 0.05,
  "lora_target_modules": [
    "q_proj",
    "v_proj",
    "k_proj",
    "out_proj",
    "fc1",
    "fc2"
  ],
  "visual_projector_hidden_dim": 1024,
  "visual_projector_dropout": 0.1,
  "use_token_type_embeddings": false,
  "use_gradient_checkpointing": false,
  "bf16": true,
  "fp16": false,
  "seed": 42
}