dataset:
  align_stage_components:
  - download/videollava/valley_llavaimage.json
  - download/videollava
  dataset_id: videollava
  dataset_root_dir: data
  finetune_stage_components:
  - download/videollava/videochatgpt_llavaimage_tune.json
  - download/videollava
  type: videollava
dataset_class: FinetuneVideoDataset
hf_token: .hf_token
model:
  align_epochs: 1
  align_global_batch_size: 256
  align_learning_rate: 0.001
  align_lr_scheduler_type: linear-warmup+cosine-decay
  align_max_grad_norm: 1.0
  align_max_steps: null
  align_per_device_batch_size: 16
  align_train_strategy: fsdp-full-shard
  align_warmup_ratio: 0.03
  align_weight_decay: 0.0
  arch_specifier: 3davg+linear
  enable_gradient_checkpointing: true
  enable_mixed_precision_training: true
  feature_fusion: cross_attention_avg_lq
  finetune_epochs: 1
  finetune_global_batch_size: 128
  finetune_learning_rate: 2.0e-05
  finetune_lr_scheduler_type: linear-warmup+cosine-decay
  finetune_max_grad_norm: 1.0
  finetune_max_steps: null
  finetune_per_device_batch_size: 8
  finetune_train_strategy: fsdp-full-shard
  finetune_warmup_ratio: 0.03
  finetune_weight_decay: 0.1
  image_resize_strategy: resize-naive
  llm_backbone_id: llama2-7b-pure
  llm_max_length: 2048
  model_id: prism-siglip-controlled+7b
  num_frames:
  - 16
  - 16
  - 32
  - 16
  projector_token_length: 64
  reduce_in_full_precision: false
  type: merv-full
  video_backbone_ids:
  - languagebind-video-noclass
  - dinov2-video-all-tokens
  - vivit-google-b-all-no-cls-16frames
  - siglip-vit-b16-224px-all-no-cls
  visual_feature_length: 1024
pretrained_checkpoint: runs/final-4e-stage1-lr1e-4/checkpoints/latest-checkpoint.pt
run_id: final-4e-stage2-(lr1e-4)
run_root_dir: runs
seed: 7
slurm_id: '751580'
stage: second_finetune
trackers:
- jsonl
- wandb