Saving weights and logs of step 10000

events.out.tfevents.1641904240.t1v-n-358ff5d1-w-0.938257.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:3006ce33284cc447090efc6ba97568b83e67e98497dc6486a0c192eb48afe1de
+size 1470136

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:508b8deebc50866620656ec0433eba688aacfa19fbd9835a9559643ac7d20b5e
+oid sha256:729c1ff63f2d39cefba876c3299619b94aeb088dfc1cf8eb4e4696b081b97e56
 size 498796983

run_step3.sh

@@ -0,0 +1,26 @@
+./run_mlm_flax.py \
+    --output_dir="./" \
+    --model_type="roberta" \
+    --model_name_or_path="./" \
+    --config_name="./" \
+    --tokenizer_name="./" \
+    --dataset_name="NbAiLab/NCC" \
+    --cache_dir="/mnt/disks/flaxdisk/cache/" \
+    --max_seq_length="128" \
+    --weight_decay="0.01" \
+    --per_device_train_batch_size="200" \
+    --per_device_eval_batch_size="200" \
+    --learning_rate="4e-4" \
+    --warmup_steps="10000" \
+    --overwrite_output_dir \
+    --num_train_epochs="10" \
+    --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

run_step4.sh

@@ -0,0 +1,26 @@
+./run_mlm_flax.py \
+    --output_dir="./" \
+    --model_type="roberta" \
+    --model_name_or_path="./" \
+    --config_name="./" \
+    --tokenizer_name="./" \
+    --dataset_name="NbAiLab/NCC" \
+    --cache_dir="/mnt/disks/flaxdisk/cache/" \
+    --max_seq_length="128" \
+    --weight_decay="0.01" \
+    --per_device_train_batch_size="200" \
+    --per_device_eval_batch_size="200" \
+    --learning_rate="2e-4" \
+    --warmup_steps="10000" \
+    --overwrite_output_dir \
+    --num_train_epochs="10" \
+    --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