Saving weights and logs of step 10000

events.out.tfevents.1641221745.t1v-n-e1a08808-w-0.680362.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:be815f4aa8dc41aa3c254c75856970ee861ed86ea363c21761b49581a6f67b19
+size 1470136

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d21af5655cba53e0c9f066b90018126ba0f14783476eca376ee8b97585be99ab
+oid sha256:c07aa650bf228dfb2376a5bc921155492f9a4924bfb37ef8e8a316c006c10ad7
 size 498796983

run_step1.sh

@@ -8,8 +8,8 @@
     --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" \
+    --per_device_train_batch_size="192" \
+    --per_device_eval_batch_size="192" \
     --learning_rate="4e-4" \
     --warmup_steps="10000" \
     --overwrite_output_dir \