test_configs/mistral_test_config.yaml

model:
  arch: mini_gpt4_llama_v2
  freeze_vit: True
  freeze_qformer: True
  max_txt_len: 384
  low_resource: True
  image_size: 224
  end_sym: "</s>"
  llama_model: "mistralai/Mistral-7B-Instruct-v0.2"
  ckpt: "checkpoints/video_mistral_checkpoint_last.pth"
  use_grad_checkpoint: True
  chat_template: True
  lora_r: 64
  lora_alpha: 16
  length: 32
  use_grad_checkpoint_llm: True
  max_context_len: 4096
  architectures: [
    "MiniGPT4_Video"
  ]
  device: "cuda"
  drop_path_rate: 0 
  img_size: 224
  model_type: "minigpt4_video"
  num_query_token: 24
  prompt: ""
  torch_dtype: "float16"
  transformers_version: "4.42.3"
  vit_precision: "fp16"
  vit_model: "eva_clip_g"
  token_pooling: true
  lora_target_modules : ["q_proj","v_proj"]
  lora_dropout: 0.05
  remove_template: false
  prompt_path: ""
  minigpt4_gpu_id: 0
  whisper_gpu_id: 0
  answer_module_gpu_id: 0
  


datasets:
  video_chatgpt: #99378 row  - 13224 video
    batch_size: 1
    vis_processor:
      train:
        name: "blip2_image_train"
        image_size: 224
    text_processor:
      train:
        name: "blip_caption"
    sample_ratio: 200


run:
  task: image_text_pretrain
  seed: 42
  amp: True