Saving weights and logs of step 10000

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": ".",
+  "_name_or_path": "./",
   "architectures": [
     "RobertaForMaskedLM"
   ],

events.out.tfevents.1642077230.t1v-n-ccbf3e94-w-0.1162950.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0e33670c3598dbcaf5f4429b3aab3dfd11efa65bec659418c6cca4da7c81594a
+size 1470136

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:cacdb8e4c2e09bba546c578d574490e5e2e80b8dbc8a29c2a35cb746bfc1b9d6
+oid sha256:9b0d23f5359aeb972fe292ae9e2d3485e2071e94d15b5be5ea1ac71c3ead79ed
 size 498796983

run_step4.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_4_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="5e-5" \
+    --end_learning_rate="0" \
+    --warmup_steps="0" \
+    --overwrite_output_dir \
+    --num_train_epochs="4" \
+    --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" \
+    --dtype="bfloat16" \
+    --push_to_hub