feat: update configs

tools/train/config/medium/config.json

@@ -4,17 +4,17 @@
   "attention_dropout": 0.0,
   "bos_token_id": 16385,
   "classifier_dropout": 0.0,
-  "d_model": 1536,
+  "d_model": 1408,
   "decoder_attention_heads": 16,
   "decoder_ffn_dim": 4096,
   "decoder_layerdrop": 0.0,
-  "decoder_layers": 18,
+  "decoder_layers": 14,
   "decoder_start_token_id": 16384,
-  "dropout": 0.1,
+  "dropout": 0.0,
   "encoder_attention_heads": 16,
   "encoder_ffn_dim": 4096,
   "encoder_layerdrop": 0.0,
-  "encoder_layers": 18,
+  "encoder_layers": 14,
   "encoder_vocab_size": 50264,
   "eos_token_id": 16385,
   "gradient_checkpointing": false,

tools/train/config/micro/config.json

@@ -8,13 +8,13 @@
   "decoder_attention_heads": 16,
   "decoder_ffn_dim": 2048,
   "decoder_layerdrop": 0.0,
-  "decoder_layers": 6,
+  "decoder_layers": 2,
   "decoder_start_token_id": 16384,
-  "dropout": 0.1,
+  "dropout": 0.0,
   "encoder_attention_heads": 16,
   "encoder_ffn_dim": 2048,
   "encoder_layerdrop": 0.0,
-  "encoder_layers": 6,
+  "encoder_layers": 2,
   "encoder_vocab_size": 50264,
   "eos_token_id": 16385,
   "gradient_checkpointing": false,

tools/train/config/mini/config.json

@@ -10,7 +10,7 @@
   "decoder_layerdrop": 0.0,
   "decoder_layers": 12,
   "decoder_start_token_id": 16384,
-  "dropout": 0.1,
+  "dropout": 0.0,
   "encoder_attention_heads": 16,
   "encoder_ffn_dim": 4096,
   "encoder_layerdrop": 0.0,