pytorch

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": "./",
+  "_name_or_path": ".",
   "architectures": [
     "RobertaForMaskedLM"
   ],
@@ -20,7 +20,7 @@
   "num_hidden_layers": 12,
   "pad_token_id": 1,
   "position_embedding_type": "absolute",
-  "torch_dtype": "bfloat16",
+  "torch_dtype": "float32",
   "transformers_version": "4.15.0.dev0",
   "type_vocab_size": 1,
   "use_cache": true,

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:65af654c9d069514d0ab3f9001e8addb8403c4ed1038ae183a874d027eaf1a21
+size 498872555

run_step2.sh

@@ -0,0 +1,28 @@
+./run_mlm_flax.py \
+    --output_dir="./" \
+    --model_type="roberta" \
+    --model_name_or_path="./" \
+    --config_name="./" \
+    --tokenizer_name="./" \
+    --train_file /mnt/disks/flaxdisk/corpus/train_2_4.json \
+    --validation_file /mnt/disks/flaxdisk/corpus/validation.json \
+    --cache_dir="/mnt/disks/flaxdisk/cache/" \
+    --max_seq_length="512" \
+    --weight_decay="0.01" \
+    --per_device_train_batch_size="40" \
+    --per_device_eval_batch_size="40" \
+    --learning_rate="2e-4" \
+    --warmup_steps="0" \
+    --overwrite_output_dir \
+    --num_train_epochs="2" \
+    --adam_beta1="0.9" \
+    --adam_beta2="0.98" \
+    --adam_epsilon="1e-6" \
+    --logging_steps="10000" \
+    --save_steps="10000" \
+    --eval_steps="10000" \
+    --preprocessing_num_workers="64" \
+    --auth_token="True" \
+    --static_learning_rate="True" \
+    --dtype="bfloat16" \
+    --push_to_hub