final

events.out.tfevents.1667893477.t1v-n-101cf975-w-0.109623.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:a029b66c294bc2a41f6155d4c046d0c5ed8d84a735518468c5f9e70de47152e9
+size 43927850

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e4de7fc184dcf3cc1b013651a25806d1417baa8be6140839cd6bdb58b3f0d2bb
+oid sha256:9042b69e0b1c94c7114eb1053998800b02dd4de191ce68ecaa7aaf5ae4c648ff
 size 1113187999

run_recover_200.sh

@@ -0,0 +1,21 @@
+python run_mlm_flax_stream.py \
+    --output_dir="../roberta-base-exp-8" \
+    --model_name_or_path="./" \
+    --config_name="./" \
+    --tokenizer_name="./" \
+    --dataset_name="NbAiLab/scandinavian" \
+    --max_seq_length="512" \
+    --weight_decay="0.01" \
+    --per_device_train_batch_size="62" \
+    --per_device_eval_batch_size="62" \
+    --learning_rate="3e-5" \
+    --warmup_steps="0" \
+    --overwrite_output_dir \
+    --num_train_steps="300000" \
+    --adam_beta1="0.9" \
+    --adam_beta2="0.98" \
+    --logging_steps="5000" \
+    --save_steps="25000" \
+    --eval_steps="25000" \
+    --dtype="bfloat16" \
+    --push_to_hub