config_cuda_100k.yml

RUN:
  DATA: shakespeare.txt
  CUDA: yes
  OUTPUT_DIR: ./100k
MODEL:
  CLASS: Transformer
  PARAMS:
    CONTEXT_SIZE: 256
    EMBEDDING_SIZE: 384
    NUM_HEADS: 6
    NUM_BLOCKS: 6
    DROPOUT_RATIO: 0.2
TRAIN:
  NUM_ITERS: 100000
  LEARNING_RATE: 1e-3
  LR_SCHEDULE: no
  BATCH_SIZE: 64
  CHECKPOINT_ITERS: 5000
EVAL:
  CADENCE_ITERS: 250
  NUM_SAMPLES: 50
  GENERATION_ITERS:
    [
      1000,
      2000,
      5000,
      10000,
      20000,
      30000,
      40000,
      50000,
      60000,
      70000,
      80000,
      90000,
      100000,
    ]
  GENERATION_SIZE: 2000