Saving weights and logs of step 1000

eval_results.json

@@ -0,0 +1,5 @@
+{
+    "eval_accuracy": 0.6890367604912527,
+    "eval_loss": 1.4896034919686423,
+    "eval_perplexity": 4.435336523745398
+}

events.out.tfevents.1643733579.t1v-n-ccbf3e94-w-0.346519.3.v2

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4fb95c57b0658fe85bfb8af4a847fde678c03be588e35de696bf848a834297b1
+size 147136

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c585c1a1ca1ec3d7d95a81312f971c41ae71e89071b6a4672221b0bb6718ac65
+oid sha256:92681f621e5c1ece7bf9ff798e84c7c86fb171b010f1026e2b651e213d16fb28
 size 498796983

run_128_recover_7e.sh

@@ -0,0 +1,28 @@
+python run_mlm_flax.py \
+    --output_dir="./" \
+    --model_type="roberta" \
+    --model_name_or_path="./" \
+    --config_name="roberta-base" \
+    --tokenizer_name="NbAiLab/nb-roberta-base" \
+    --dataset_name="NbAiLab/NCC" \
+    --cache_dir="/mnt/disks/flaxdisk/cache/" \
+    --max_seq_length="128" \
+    --weight_decay="0.01" \
+    --per_device_train_batch_size="232" \
+    --per_device_eval_batch_size="232" \
+    --pad_to_max_length \
+    --learning_rate="1.8183001770921692e-07" \
+    --warmup_steps="0" \
+    --overwrite_output_dir \
+    --num_train_epochs="3" \
+    --adam_beta1="0.9" \
+    --adam_beta2="0.98" \
+    --adam_epsilon="1e-6" \
+    --logging_steps="1000" \
+    --save_steps="1000" \
+    --eval_steps="1000" \
+    --auth_token="True" \
+    --do_train \
+    --do_eval \
+    --dtype="bfloat16" \
+    --push_to_hub