lavis/projects/alpro/train/msrvtt_qa_ft.yaml

 # Copyright (c) 2022, salesforce.com, inc.
 # All rights reserved.
 # SPDX-License-Identifier: BSD-3-Clause
 # For full license text, see the LICENSE file in the repo root or https://opensource.org/licenses/BSD-3-Clause

model:
  arch: alpro_qa
  model_type: msrvtt
  load_finetuned: False

  num_classes: 1500

  timesformer:
    use_grad_ckpt: True
    ckpt_layer: 12


datasets:
  msrvtt_qa: # name of the dataset builder
    vis_processor:
        train:
          name: "alpro_video_train"
          n_frms: 16
          image_size: 224
        eval:
          name: "alpro_video_eval"
          n_frms: 16
          image_size: 224
    text_processor:
        train:
          name: "blip_caption"
        eval:
          name: "blip_caption"

run:
  task: multimodal_classification
  # optimization-specific
  lr_sched: "linear_warmup_cosine_lr"
  init_lr: 5e-5
  min_lr: 1e-6
  weight_decay: 1e-4
  max_epoch: 10
  batch_size_train: 24
  batch_size_eval: 64
  num_workers: 4

  seed: 42
  output_dir: "output/ALPRO/msrvtt_qa"

  amp: False
  resume_ckpt_path: null

  evaluate: False 
  train_splits: ["train"]
  valid_splits: ["val", "test"]
  test_splits: ["test"]

  # distribution-specific
  device: "cuda"
  world_size: 1
  dist_url: "env://"
  distributed: True