Saving weights and logs of step 10000

events.out.tfevents.1629902662.t1v-n-1ae8dadb-w-0.181842.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:830bfc14ddfb9e53f03a7c79030b4fc5ddee472ee809578fca2e3d227a8cc3c7
+size 1470757

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:314d30f9c83fc85ad87e87e0d996870c813bd69c8c04678a1aadeae47b00397c
+oid sha256:9c8274c4dd16c72c5f36f1e4441851a0dd44dd2f1fe464977bbf43174e6b6a12
 size 711588089

start_train.sh

@@ -10,8 +10,8 @@ python3 run_mlm_flax.py \
     --max_seq_length="128" \
     --pad_to_max_length \
     --preprocessing_num_workers="96" \
-    --per_device_train_batch_size="128" \
-    --per_device_eval_batch_size="128" \
+    --per_device_train_batch_size="64" \
+    --per_device_eval_batch_size="64" \
     --learning_rate="3e-4" \
     --warmup_steps="1000" \
     --overwrite_output_dir \
@@ -22,5 +22,4 @@ python3 run_mlm_flax.py \
     --eval_steps="10000" \
     --logging_steps="1000" \
     --dtype="bfloat16" \
-    --adafactor \
     --push_to_hub