Change default finetune model to the plus version

Use --group_by_length to speed up training

Author: czczup

internvl_chat/evaluate.sh

@@ -35,9 +35,9 @@ if  [ ${DATASET} == "caption-coco" ]; then
   torchrun \
     --nnodes=1 \
     --node_rank=0 \
-    --master_addr=127--master_port=${MASTER_PORT}.0.0.1 \
+    --master_addr=127.0.0.1 \
     --nproc_per_node=${GPUS} \
-     \
+    --master_port=${MASTER_PORT} \
     eval/caption/evaluate_caption.py --checkpoint ${CHECKPOINT} --datasets coco
 fi
 

internvl_chat/shell/hermes2_yi34b/internvl_chat_v1_2_hermes2_yi34b_448_finetune.sh

@@ -67,6 +67,7 @@ srun -p ${PARTITION} \
   --lr_scheduler_type "cosine" \
   --logging_steps 1 \
   --max_seq_length 2048 \
+  --group_by_length True \
   --do_train True \
   --grad_checkpoint True \
   --deepspeed "zero_stage3_config.json" \

internvl_chat/shell/hermes2_yi34b/internvl_chat_v1_2_hermes2_yi34b_448_finetune_continue.sh

@@ -36,7 +36,7 @@ srun -p ${PARTITION} \
   --quotatype=${QUOTA_TYPE} \
   ${SRUN_ARGS} \
   python -u internvl/train/internvl_chat_finetune.py \
-  --model_name_or_path "./pretrained/InternVL-Chat-Chinese-V1-2" \
+  --model_name_or_path "./pretrained/InternVL-Chat-Chinese-V1-2-Plus" \
   --conv_style "Hermes-2" \
   --output_dir ${OUTPUT_DIR} \
   --meta_path "./path/to/your/custom/meta/file" \
@@ -65,6 +65,7 @@ srun -p ${PARTITION} \
   --lr_scheduler_type "cosine" \
   --logging_steps 1 \
   --max_seq_length 2048 \
+  --group_by_length True \
   --do_train True \
   --grad_checkpoint True \
   --deepspeed "zero_stage3_config.json" \

internvl_chat/shell/hermes2_yi34b/internvl_chat_v1_2_hermes2_yi34b_448_finetune_continue_lora.sh

@@ -29,7 +29,7 @@ torchrun \
   --master_port=${MASTER_PORT} \
   ${SRUN_ARGS} \
   internvl/train/internvl_chat_finetune.py \
-  --model_name_or_path "./pretrained/InternVL-Chat-Chinese-V1-2" \
+  --model_name_or_path "./pretrained/InternVL-Chat-Chinese-V1-2-Plus" \
   --conv_style "Hermes-2" \
   --output_dir ${OUTPUT_DIR} \
   --meta_path "./path/to/your/custom/meta/file" \
@@ -59,6 +59,7 @@ torchrun \
   --lr_scheduler_type "cosine" \
   --logging_steps 1 \
   --max_seq_length 2048 \
+  --group_by_length True \
   --do_train True \
   --grad_checkpoint True \
   --deepspeed "zero_stage3_config.json" \