Saving weights and logs of step 1000

events.out.tfevents.1644752234.t1v-n-79f0077b-w-0.16365.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ba7a603585bf059429e046e56fe9e436fc8ec750fb0c8f249638ed575babf108
+size 147136

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e87f2bfbc92fd8b3d74253235448f92a50dd2654de6698969fcab07d0afb1018
+oid sha256:c8de9d66d0e8008a170089cefec2ee882d37cbb8f55284e50a01b61216ac3c12
 size 498796983

run_128_scandinavian_recover100k.sh

@@ -0,0 +1,32 @@
+#    --per_device_train_batch_size="232" \
+#    --per_device_eval_batch_size="232" \
+
+
+python run_mlm_flax.py \
+    --output_dir="./" \
+    --model_type="roberta" \
+    --model_name_or_path="./" \
+    --config_name="roberta-base" \
+    --tokenizer_name="NbAiLab/nb-roberta-base" \
+    --dataset_name="NbAiLab/scandinavian" \
+    --cache_dir="/mnt/disks/flaxdisk/cache/" \
+    --max_seq_length="128" \
+    --weight_decay="0.01" \
+    --per_device_train_batch_size="116" \
+    --per_device_eval_batch_size="116" \
+    --pad_to_max_length \
+    --learning_rate="2e-4" \
+    --warmup_steps="10000" \
+    --overwrite_output_dir \
+    --num_train_epochs="9" \
+    --adam_beta1="0.9" \
+    --adam_beta2="0.98" \
+    --adam_epsilon="1e-6" \
+    --logging_steps="1000" \
+    --save_steps="1000" \
+    --eval_steps="1000" \
+    --auth_token="True" \
+    --do_train \
+    --do_eval \
+    --dtype="bfloat16" \
+    --push_to_hub