Training in progress, step 365

config.json

@@ -3,9 +3,9 @@
     "T5ForConditionalGeneration"
   ],
   "classifier_dropout": 0.0,
-  "d_ff": 1024,
+  "d_ff": 2048,
   "d_kv": 64,
-  "d_model": 512,
+  "d_model": 768,
   "decoder_start_token_id": 0,
   "dense_act_fn": "gelu_new",
   "dropout_rate": 0.1,
@@ -18,9 +18,9 @@
   "layer_norm_epsilon": 1e-06,
   "model_type": "t5",
   "n_positions": 512,
-  "num_decoder_layers": 8,
-  "num_heads": 6,
-  "num_layers": 8,
+  "num_decoder_layers": 12,
+  "num_heads": 12,
+  "num_layers": 12,
   "output_past": true,
   "pad_token_id": 0,
   "relative_attention_max_distance": 128,

model.safetensors

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f201774b35dcf41762f28b0599b696b72f6bd07ae2ca16dac3ff7f631512ebcc
-size 307867048
+oid sha256:d06ef009ff667d677c05c85095d32c283bd6559fdd2e41c2a68a251e561f892d
+size 990345064

training_args.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:077b812859d3350666e80264a1056576480c069c416e67073260e7b0f8ca276d
+oid sha256:600b8ae5c2df9a2cbb061b7bcad04e5ea3f7edfd6965442e2e30d0500813fa57
 size 5905