# TAO video_clip model block for this release.
# model.safetensors (LoRA merged): build VideoCLIPPlModel with this config as-is (peft disabled) and load_state_dict(strict=True).
# model.pth (LoRA adapters unmerged, Lightning checkpoint): set peft.enabled: true and peft.vision.enabled: true, then
#   VideoCLIPPlModel.load_from_checkpoint('model.pth', experiment_spec=cfg).
# model.text_encoder = Apple MobileCLIP-B-LT weights (mobileclip_blt.pt); base vision/alignment weights resolve from
# model.internvideo2clip_hf_id through the Hugging Face cache.
model:
  type: internvideo2-clip-l14
  adaptor_name: null
  freeze_vision_encoder: true
  freeze_text_encoder: true
  image_size: 224
  init_logit_scale: 4.605170185988092
  init_logit_bias: 0.0
  canonicalize_text: false
  pretrained_ckpt: null
  internvideo2clip_hf_id: OpenGVLab/InternVideo2_distillation_models
  vision_encoder: null
  text_encoder: /models/mobileclip_blt.pt
  clip_head: null
  num_frames: 8
  use_flash_attn: false
  use_fused_rmsnorm: false
  use_fused_mlp: false
peft:
  enabled: false
  method: lora
  vision:
    enabled: false
    target_modules:
    - qkv
    - proj
    num_last_blocks: 12
    rank: 8
    alpha: 16
    dropout: 0.05
  text:
    enabled: false
    target_modules:
    - qkv_proj
    - out_proj
    num_last_blocks: 3
    rank: 8
    alpha: 16
    dropout: 0.05
regularization:
  enabled: false