update

examples/train/multimodal/infer.sh

@@ -0,0 +1,6 @@
+# Perform inference using the validation set from the training phase.
+# CUDA_VISIBLE_DEVICES=0,1 \
+swift infer \
+    --adapters output/vx-xxx/checkpoint-xxx \
+    --stream true \
+    --load_data_args true

examples/train/multimodal/ocr.sh

@@ -0,0 +1,25 @@
+# 2*75GB
+CUDA_VISIBLE_DEVICES=0,1 \
+MAX_PIXELS=1003520 \
+swift sft \
+    --model Qwen/QVQ-72B-Preview \
+    --dataset AI-ModelScope/LaTeX_OCR:human_handwrite#20000 \
+    --train_type lora \
+    --torch_dtype bfloat16 \
+    --num_train_epochs 1 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
+    --learning_rate 1e-4 \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --freeze_vit true \
+    --gradient_accumulation_steps 16 \
+    --eval_steps 50 \
+    --save_steps 50 \
+    --save_total_limit 5 \
+    --logging_steps 5 \
+    --max_length 2048 \
+    --output_dir output \
+    --warmup_ratio 0.05 \
+    --dataloader_num_workers 4

examples/train/multimodal/video.sh

@@ -0,0 +1,29 @@
+# 4*80GB
+# You can refer to `https://github.com/QwenLM/Qwen2-VL` for the meaning of the `VIDEO_MAX_PIXELS` parameter.
+NPROC_PER_NODE=4 \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+VIDEO_MAX_PIXELS=200704 \
+FPS_MAX_FRAMES=24 \
+swift sft \
+    --model Qwen/QVQ-72B-Preview \
+    --dataset swift/VideoChatGPT:all \
+    --train_type lora \
+    --torch_dtype bfloat16 \
+    --num_train_epochs 1 \
+    --per_device_train_batch_size 1 \
+    --per_device_eval_batch_size 1 \
+    --learning_rate 1e-4 \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --target_modules all-linear \
+    --freeze_vit true \
+    --gradient_accumulation_steps 4 \
+    --eval_steps 50 \
+    --save_steps 50 \
+    --save_total_limit 5 \
+    --logging_steps 5 \
+    --max_length 2048 \
+    --output_dir output \
+    --warmup_ratio 0.05 \
+    --dataloader_num_workers 4 \
+    --deepspeed zero3