continued 111001 training steps

continue_run_mlm_flax_stream.sh

@@ -1,22 +1,22 @@
-export MODEL_DIR=/home/Z6HJB/roberta-large-scandi/roberta-base-scandinavian/
+export MODEL_DIR=/home/Z6HJB/train-roberta-base-scandinavian/roberta-base-scandinavian/
 
 source /home/Z6HJB/test/bin/activate
 
-python3 ./src/run_mlm_flax_stream.py \
+python3 ./src/scandinavian_run_mlm_flax_stream.py \
     --model_name_or_path="${MODEL_DIR}" \
     --output_dir="${MODEL_DIR}" \
     --tokenizer_name="${MODEL_DIR}" \
     --dataset_name="mc4" \
     --dataset_config_name="unshuffled_deduplicated_en" \
-    --max_seq_length="514" \
-    --per_device_train_batch_size="32" \
-    --per_device_eval_batch_size="32" \
-    --learning_rate="3e-4" \
+    --max_seq_length="128" \
+    --per_device_train_batch_size="128" \
+    --per_device_eval_batch_size="128" \
+    --learning_rate="1.087e-6" \
     --warmup_steps="1000" \
     --overwrite_output_dir \
     --adam_beta1="0.9" \
     --adam_beta2="0.98" \
-    --num_train_steps="100000" \
+    --num_train_steps="200000" \
     --num_eval_samples="5000" \
     --save_steps="1000" \
     --logging_steps="250" \

events.out.tfevents.1625999437.t1v-n-58ed2201-w-0.98932.3.v2 → events.out.tfevents.1626175943.t1v-n-58ed2201-w-0.155891.3.v2

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebb0542eea19cce6e79e62a7fd7ad47ac611d2c6207b8003106d5c297a23f7b7
-size 10389142
+oid sha256:8ffed2c9c5c4999da79f6278e1e5747558ffbb3f39f236eb56271bd130c5447b
+size 16627546

flax_model.msgpack

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8c12e775fc0be8428781ba4bbee9c613fbc8bd414fb1d80fdf5497c24ba7ca0d
+oid sha256:7ef047be1c30de5fd9b499f1d217091eb0baf194be896c002d4c2e87939639c7
 size 498796983