merge data part to the text stream

Former-commit-id: 7ee20286d9bcc2d5378bfd6bb02cd3648396d873

examples/mllm/sft_instructblip.sh

@@ -1,32 +0,0 @@
-#!/bin/bash
-
-CUDA_VISIBLE_DEVICES=0 python src/train_bash.py \
-    --stage sft_mm \
-    --do_train \
-    --model_name_or_path Salesforce/instructblip-vicuna-7b \
-    --dataset mllm_instruct_example \
-    --dataset_dir data \
-    --template default \
-    --finetuning_type lora \
-    --lora_target all \
-    --output_dir saves/instructblip-vicuna-7b/lora/sft \
-    --overwrite_cache \
-    --overwrite_output_dir \
-    --cutoff_len 1024 \
-    --preprocessing_num_workers 16 \
-    --per_device_train_batch_size 3 \
-    --per_device_eval_batch_size 1 \
-    --gradient_accumulation_steps 1 \
-    --lr_scheduler_type cosine \
-    --logging_steps 1 \
-    --warmup_steps 20 \
-    --save_steps 100 \
-    --eval_steps 100 \
-    --evaluation_strategy steps \
-    --load_best_model_at_end \
-    --learning_rate 1e-5 \
-    --num_train_epochs 50 \
-    --max_samples 3000 \
-    --val_size 0.1 \
-    --plot_loss \
-    --bf16