Saving weights and logs of step 5000

config.json

@@ -1,5 +1,5 @@
 {
-  "_name_or_path": ".",
+  "_name_or_path": "./",
   "architectures": [
     "RobertaForMaskedLM"
   ],

events.out.tfevents.1639431874.t1v-n-ccbf3e94-w-0.1053519.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:68887780af1577a229cee0ce20bb9da8be2c8465ca4f2b6508402c6256754f4a
+size 735136

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a993bcaf3ce549644ccbe6e0640cf03169148e4fbf4700be812d6e5a0b11164
+oid sha256:9ff2cd2aa2ccf0aa8ecc9af3e713ed51b9f3814590547b10d550c380ee3bc0df
 size 498796983

run_w5_long.sh

@@ -3,8 +3,9 @@
     --model_type="roberta" \
     --config_name="./" \
     --tokenizer_name="./" \
-    --train_file="/mnt/disks/flaxdisk/corpus/train-shard-0001-of-0001.json" \
+    --train_file="/mnt/disks/flaxdisk/corpus/long_train.json" \
     --validation_file="/mnt/disks/flaxdisk/corpus/validation-shard-0001-of-0001.json" \
+    --cache_dir="/mnt/disks/flaxdisk/cache/" \
     --max_seq_length="128" \
     --weight_decay="0.01" \
     --per_device_train_batch_size="128" \
@@ -18,4 +19,5 @@
     --logging_steps="5000" \
     --save_steps="5000" \
     --eval_steps="5000" \
+    --preprocessing_num_workers="64" \
     --push_to_hub