even smaller batch size

finetune_large_mt5_sentencefix_v4_16.gin

@@ -23,7 +23,7 @@ RANDOM_SEED = 0
 # `2048 * 114`. For mT5: `1024 * 229`. For ByT5: `1024 * 189`.
 #LOSS_NORMALIZING_FACTOR = 234496
 INITIAL_CHECKPOINT_PATH = "gs://t5-data/pretrained_models/t5x/mt5_large/checkpoint_1000000"
-BATCH_SIZE = 512
+BATCH_SIZE = 256
 
 #train_script.train:
 #  eval_period = 500