big transformer

config.json

@@ -1,26 +1,26 @@
 {
   "_name_or_path": "bart-base",
-  "activation_dropout": 0.1,
+  "activation_dropout": 0.3,
   "activation_function": "gelu",
   "add_bias_logits": false,
   "add_final_layer_norm": false,
   "architectures": [
     "BartModel"
   ],
-  "attention_dropout": 0.1,
+  "attention_dropout": 0.3,
   "bos_token_id": 0,
-  "classif_dropout": 0.1,
+  "classif_dropout": 0.3,
   "classifier_dropout": 0.0,
-  "d_model": 512,
-  "decoder_attention_heads": 8,
-  "decoder_ffn_dim": 2048,
+  "d_model": 1024,
+  "decoder_attention_heads": 16,
+  "decoder_ffn_dim": 4096,
   "decoder_layerdrop": 0.0,
   "decoder_layers": 6,
   "decoder_start_token_id": 2,
-  "dropout": 0.1,
+  "dropout": 0.3,
   "early_stopping": true,
-  "encoder_attention_heads": 8,
-  "encoder_ffn_dim": 2048,
+  "encoder_attention_heads": 16,
+  "encoder_ffn_dim": 4096,
   "encoder_layerdrop": 0.0,
   "encoder_layers": 6,
   "eos_token_id": 2,