feat: split script for small and big runs

seq2seq/do_big_run.sh

@@ -0,0 +1,16 @@
+python run_seq2seq_flax.py \
+	--max_source_length 128 \
+	--train_file /data/CC12M/encoded-small-train.tsv \			# ignored for now in our script
+	--validation_file /data/CC12M/encoded-small-valid.tsv \		# ignored for now in our script
+	--output_dir output \
+	--per_device_train_batch_size 56 \
+	--per_device_eval_batch_size 56 \
+	--preprocessing_num_workers 80 \
+	--warmup_steps 125 \
+	--gradient_accumulation_steps 8 \
+	--do_train \
+	--do_eval \
+	--adafactor \
+	--num_train_epochs 10 \
+	--log_model \
+	--learning_rate 0.001

seq2seq/{do_run.sh → do_small_run.sh}

@@ -1,7 +1,7 @@
 python run_seq2seq_flax.py \
 	--max_source_length 128 \
-	--train_file /data/CC12M/encoded-small-train.tsv \
-	--validation_file /data/CC12M/encoded-small-valid.tsv \
+	--train_file /data/CC12M/encoded-small-train.tsv \			# ignored for now in our script
+	--validation_file /data/CC12M/encoded-small-valid.tsv \		# ignored for now in our script
 	--output_dir output \
 	--per_device_train_batch_size 56 \
 	--per_device_eval_batch_size 56 \