Saving weights and logs of step 10000

events.out.tfevents.1640956682.t1v-n-358ff5d1-w-0.11974.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8b0fe1349611f01cd08d8a11929610ed03a8c20073c1fa070a7b7843955235bf
+size 1470136

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:5a7c59a989095aa103c94d09f11a9b5cd017f6fc2e3b4603db111c5b9d8bfa42
+oid sha256:cc49f760bccb82672d93ae5c122aed1f410add84860109051c07d287d42a5517
 size 498796983

run_step1.sh

@@ -7,12 +7,12 @@
     --cache_dir="/mnt/disks/flaxdisk/cache/" \
     --max_seq_length="128" \
     --weight_decay="0.01" \
-    --per_device_train_batch_size="164" \
-    --per_device_eval_batch_size="164" \
-    --learning_rate="3e-4" \
+    --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="1000" \
+    --num_train_epochs="10" \
     --adam_beta1="0.9" \
     --adam_beta2="0.98" \
     --adam_epsilon="1e-6" \