Update finetune_ds.sh

finetune/finetune_ds.sh

@@ -6,11 +6,11 @@ NODE_RANK=0
 MASTER_ADDR=localhost
 MASTER_PORT=6001
 
-MODEL="openbmb/MiniCPM-Llama3-V-2_5" # or openbmb/MiniCPM-V-2
+MODEL="XXXXXXX/MiniCPM-Llama3-V-2_5" # or openbmb/MiniCPM-V-2
 # ATTENTION: specify the path to your training data, which should be a json file consisting of a list of conversations.
 # See the section for finetuning in README for more information.
-DATA="path/to/trainging_data"
-EVAL_DATA="path/to/test_data"
+DATA="/home/wanglch/MiniCPM-V/data/self_build/train_data/train_data.json"
+EVAL_DATA="/home/wanglch/MiniCPM-V/data/self_build/eval_data/eval_data.json"
 LLM_TYPE="llama3" # if use openbmb/MiniCPM-V-2, please set LLM_TYPE=minicpm
 
 DISTRIBUTED_ARGS="
@@ -28,20 +28,18 @@ torchrun $DISTRIBUTED_ARGS finetune.py  \
     --remove_unused_columns false \
     --label_names "labels" \
     --prediction_loss_only false \
-    --bf16 false \
-    --bf16_full_eval false \
-    --fp16 true \
-    --fp16_full_eval true \
+    --bf16 true \
+    --bf16_full_eval true \
     --do_train \
     --do_eval \
     --tune_vision true \
     --tune_llm true \
     --model_max_length 2048 \
     --max_slice_nums 9 \
-    --max_steps 10000 \
-    --eval_steps 1000 \
-    --output_dir output/output_minicpmv2 \
-    --logging_dir output/output_minicpmv2 \
+    --max_steps 100 \
+    --eval_steps 10 \
+    --output_dir  "/home/wanglch/MiniCPM-V/saves/MiniCPM-Llama3-V-2_5/train_lora/" \
+    --logging_dir  "/home/wanglch/MiniCPM-V/saves/MiniCPM-Llama3-V-2_5/train_lora/" \
     --logging_strategy "steps" \
     --per_device_train_batch_size 1 \
     --per_device_eval_batch_size 1 \
@@ -57,5 +55,5 @@ torchrun $DISTRIBUTED_ARGS finetune.py  \
     --lr_scheduler_type "cosine" \
     --logging_steps 1 \
     --gradient_checkpointing true \
-    --deepspeed ds_config_zero2.json \
+    --deepspeed ds_config_zero3.json \
     --report_to "tensorboard"