configs/BERT_L12_H768_experiments/zeroshot_config/msvd_retrieval.yaml

_BASE_: "../base_model_bert_l12_h768.yaml"


TASKS:

  -
    NAME: msvd_retrieval
    DATASETS:
      TRAIN: 'MSVDDataset'
      TEST: 'MSVDDataset'
      TASK_TYPE: 'video_retrieval'
      DATASET_NAME: 'MSVDDataset'
      # TARGET_SET: ['Vocab_Word']
    DATALOADER:
      TRAIN_BATCH_SIZE: 6
      TEST_BATCH_SIZE: 64
      NUM_WORKERS: 6
      FEATS_FOLDER: 'open_source_dataset/msvd_dataset/YouTubeClips'
      ANNO_FOLDER:  'open_source_dataset/msvd_dataset/new_annotations'
      STRIDE: 32
      FRAMES_PER_CLIP: 4
      S3_PATH: 's3://msvd/YouTubeClips/'
      TIMESFORMER_AUG: True
      SAMPLING_WEIGHT: 0.5
    MODEL:
      MAX_SEQ_LEN: 77
      EVAL_MAX_SEQ_LEN: 21
      TEMP_NAME: logit_scale_caption
    LOSSES:
      NAMES: ['LabelSmoothingCrossEntropy', 'Accuracy']
      LABELSMOOTHING: 0.1
      # NAMES: ['CrossEntropy', 'Accuracy']
      LOSS_WEIGHT: 1.0
      REDUCTION: 'mean'
    INFERENCE:
      NAME: 'RetrievalEvaler'
      GENERATION_MODE: False



ENGINE:
  NAME: 'UnifiedTrainer'
 
MODEL:
  META_ARCHITECTURE: 'MultiTaskTransformerEncoder'
  ENCODER: 'UnifiedBertEncoder'

  SHARE_LAYERNORM: True
  BERT:
    NORMALIZE_DECISION: "BERTPre" 
    DROP_PATH_PROB: 0.1
    DROP_PATH_PROB_FIXED: True


  MODEL_EMA: False
  MODEL_EMA_DECAY: 0.9999

  MAEParamsInit: True
  POSEMBEDFIX: True


  IMG_INPUT_SIZE: 224
  PATCH_SIZE: 16

  # POSEMBED_SCALE: !!python/object/apply:eval ["160/224"]
  CHECKPOINT_FILETER: False 
  OLD_CHECKPONT: True 


  LAYER_SCALE: True 
  LAYER_SCALE_INIT: 1e-3


DATALOADER:
  USE_WEIGHTED_SAMPLER: True
  UNIFIED_DATASET: True 
  NUM_WORKERS: 6

  PADDING_TO_MAX: False # True for debugging or token moe with distributed moe 


  
####################################### Optimizer #######################################
SOLVER:
  NAME: 'Adam'
  TORCH_OPTIMIZER: True
  PARAMS_SEPERATE: True
  # PARAMS_GROUP: True
  # EPOCH: 1
  MAX_ITER: 450000
  CHECKPOINT_PERIOD: 50000
  EVAL_PERIOD: 500000
  BASE_LR: 0.001
  BIAS_LR_FACTOR: 1.0
  WEIGHT_DECAY: 0.05
  WEIGHT_DECAY_NORM: 0.0
  WEIGHT_DECAY_BIAS: 0.0
  WEIGHT_DECAY_EMBEDDING: 0.0
  MOMENTUM: 0.9
  DAMPENING: 0.0
  NESTEROV: 0.0
  BETAS: [0.9, 0.95]
  EPS: 1e-6
  GRAD_CLIP: 0.1
  GRAD_CLIP_TYPE: 'norm'
  ACCUM_ITER: 0
  AMP_FP16: True
  APEX_FP16: False # dangerous

  WRITE_PERIOD: 50
  MIN_LOSS_SCLE: 2048.0
  # BF16: False # True
  # ZEROSTAGE: 2

  LOSS_SCALE_WINDOW: 200





  
####################################### lr scheduler ####################################### 
LR_SCHEDULER:
  NAME: 'WarmupCosine'
  WARMUP: 20000
  MIN_LR: 0.000001




####################################### evaluation ####################################### 
INFERENCE:

  VOCAB: 'CLIP'
  ITER_BASED: True


find_unused_parameters: true

# ENCODERS:
#   -
#     NAME: VisualEncoder
#     TYPE: VisualEncoder
#     DROP_PATH_PROB: 0.0
#     HIDDEN_SIZE: 192
#     HIDDEN_DROPOUT_PROB: 0.
#     HIDDEN_ACT: "gelu"
#     NUM_ATTENTION_HEADS: 3
#     INTERMEDIATE_SIZE: 768
#     INTERMEDIATE_DROP: 0.
#     FFN_DROPOUT_PROB: 0.
#     ATTENTION_PROBS_DROPOUT_PROB: 0.
#     NUM_HIDDEN_LAYERS: 6
#     NUM_GENERATION_LAYERS: 0
#     DROP_PATH_PROB_FIXED: True

#   -
#     NAME: TextEncoder
#     TYPE: TextEncoder
#     DROP_PATH_PROB: 0.0
#     HIDDEN_SIZE: 192
#     HIDDEN_DROPOUT_PROB: 0.
#     HIDDEN_ACT: "gelu"
#     NUM_ATTENTION_HEADS: 3
#     INTERMEDIATE_SIZE: 768
#     INTERMEDIATE_DROP: 0.
#     FFN_DROPOUT_PROB: 0.
#     ATTENTION_PROBS_DROPOUT_PROB: 0.
#     NUM_HIDDEN_LAYERS: 6
#     NUM_GENERATION_LAYERS: 0
#     DROP_PATH_PROB_FIXED: True