Upload 2 files

config_big.yaml

@@ -0,0 +1,30 @@
+experiment_name: "runs/transformer_big"
+
+dataset:
+  src_lang: 'lo'
+  src_tokenizer: 'BPE'
+  src_max_seq_len: 400
+  tgt_lang: 'vi'
+  tgt_tokenizer: 'WordLevel'
+  tgt_max_seq_len: 350
+  train_dataset: 'train_clean.dat'
+  validate_dataset: 'dev_clean.dat'
+  tokenizer_file: "tokenizer_{0}.json"
+
+model: # 42688527 parameters
+  d_model: 512
+  num_heads: 8
+  d_ff: 2048
+  dropout_p: 0.3
+  num_encoder_layers: 4
+  num_decoder_layers: 2
+  model_folder: "weights"
+  model_basename: "transformer_big_"
+  preload:
+
+train:
+  lr: 0.001 # 1e-2
+  batch_size: 16
+  num_epochs: 40
+  label_smoothing: 0.1
+  on_colab: True # are you training on Colab?

transformer_big.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:09d82bd3a97ddc94c74c9ef688478b3fe5f61708445c4f7f304170a243571cb5
+size 501299424