Saving weights and logs of step 1000

events.out.tfevents.1642203685.t1v-n-eedfb410-w-0.10537.0.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f5f49581e10e342c9432818c78cf138d852e9c2d91fa0ede2deaa3b6935e8a48
+size 147136

events.out.tfevents.1642204242.t1v-n-eedfb410-w-0.profile-empty

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:5ac614ecef2709e4ed2bc443ce4ade10122a22097363c5eb86dfadf8e74fa7c5
+size 40

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9c84b9a373f89c06878b97a8ac8910af787e86c3b7c2a29cbc220446531a21e2
+oid sha256:a7856df89509557d606fad1b5333e28c86c7485f2b7ad4de66c6219ca6895c54
 size 498796983

run_mlm_flax.py

@@ -553,7 +553,7 @@ def main():
             wandb.init(
                 entity='versae',
                 project='roberta-base-ncc',
-                sync_tensorboard=True,
+                sync_tensorboard=False,
             )
             wandb.config.update(training_args)
             wandb.config.update(model_args)

train.128.sh

@@ -6,8 +6,8 @@ python run_mlm_flax.py \
     --dataset_name="NbAiLab/NCC" \
     --max_seq_length="128" \
     --weight_decay="0.01" \
-    --per_device_train_batch_size="250" \
-    --per_device_eval_batch_size="250" \
+    --per_device_train_batch_size="232" \
+    --per_device_eval_batch_size="232" \
     --pad_to_max_length \
     --learning_rate="6e-4" \
     --warmup_steps="10000" \