Initial commit

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat_int8/qlora_ddp_ds/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 19GB GPU memory
+nproc_per_node=2
+
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1 \
+torchrun \
+    --nproc_per_node=$nproc_per_node \
+    --master_port 29500 \
+    llm_sft.py \
+    --model_id_or_path qwen/Qwen-7B-Chat-Int8 \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype fp16 \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset damo-agent-mini-zh \
+    --train_dataset_sample 20000 \
+    --num_train_epochs 1 \
+    --max_length 4096 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_target_modules ALL \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps $(expr 16 / $nproc_per_node) \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn false \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/full_mp/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-audio-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/full_mp/sft.sh

+# Experimental environment: 2 * A100
+# 2 * 50GB GPU memory
+CUDA_VISIBLE_DEVICES=0,1 \
+swift sft \
+    --model_type qwen-audio-chat \
+    --sft_type full \
+    --train_dataset_sample -1 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --learning_rate 1e-5 \
+    --use_flash_attn true \
+    --save_only_model true \
+    --dataset aishell1-mini-zh \
+    --lazy_tokenize true \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/full_mp_ddp/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-audio-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/full_mp_ddp/sft.sh

+# Experimental environment: 4 * A100
+# 4 * 50GB GPU memory
+NPROC_PER_NODE=2 \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+swift sft \
+    --model_type qwen-audio-chat \
+    --sft_type full \
+    --train_dataset_sample -1 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --learning_rate 1e-5 \
+    --use_flash_attn true \
+    --save_only_model true \
+    --dataset aishell1-mini-zh \
+    --lazy_tokenize true \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/lora/infer.sh

+# Experimental environment: V100, A10, 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-audio-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.3 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/lora/sft.sh

+# Experimental environment: V100, A10, 3090
+# 21GB GPU memory
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_sft.py \
+    --model_id_or_path qwen/Qwen-Audio-Chat \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --dataset aishell1-mini-zh \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_target_modules DEFAULT \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps 16 \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn false \
+    --lazy_tokenize true \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/lora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-audio-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.3 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_audio_chat/lora_ddp_ds/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 21GB GPU memory
+nproc_per_node=2
+
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1 \
+torchrun \
+    --nproc_per_node=$nproc_per_node \
+    --master_port 29500 \
+    llm_sft.py \
+    --model_type qwen-audio-chat \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset aishell1-mini-zh \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_target_modules DEFAULT \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps $(expr 16 / $nproc_per_node) \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn false \
+    --deepspeed default-zero2 \
+    --lazy_tokenize true \

swift-main/examples/pytorch/llm/scripts/qwen_vl/lora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-vl/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.7 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_vl/lora_ddp_ds/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 21GB GPU memory
+nproc_per_node=2
+
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1 \
+torchrun \
+    --nproc_per_node=$nproc_per_node \
+    --master_port 29500 \
+    llm_sft.py \
+    --model_id_or_path qwen/Qwen-VL \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type default \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset coco-en-mini \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_target_modules c_attn attn.c_proj \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps $(expr 16 / $nproc_per_node) \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn false \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/full_mp/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-vl-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/full_mp/sft.sh

+# Experimental environment: 2 * A100
+# 2 * 55GB GPU memory
+CUDA_VISIBLE_DEVICES=0,1 \
+swift sft \
+    --model_type qwen-vl-chat \
+    --sft_type full \
+    --train_dataset_sample -1 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --learning_rate 1e-5 \
+    --use_flash_attn true \
+    --save_only_model true \
+    --dataset coco-en-mini \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/full_mp_ddp/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-vl-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/full_mp_ddp/sft.sh

+# Experimental environment: 4 * A100
+# 4 * 55GB GPU memory
+NPROC_PER_NODE=2 \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+swift sft \
+    --model_type qwen-vl-chat \
+    --sft_type full \
+    --train_dataset_sample -1 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --learning_rate 1e-5 \
+    --use_flash_attn true \
+    --save_only_model true \
+    --dataset coco-en-mini \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/lora/infer.sh

+# Experimental environment: V100, A10, 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-vl-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.7 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/lora/sft.sh

+# Experimental environment: V100, A10, 3090
+# 21GB GPU memory
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_sft.py \
+    --model_id_or_path qwen/Qwen-VL-Chat \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --dataset coco-en-mini \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_target_modules DEFAULT \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps 16 \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn false \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/lora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-vl-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.7 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/lora_ddp_ds/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 21GB GPU memory
+nproc_per_node=2
+
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1 \
+torchrun \
+    --nproc_per_node=$nproc_per_node \
+    --master_port 29500 \
+    llm_sft.py \
+    --model_id_or_path qwen/Qwen-VL-Chat \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset coco-en-mini \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_target_modules c_attn attn.c_proj \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps $(expr 16 / $nproc_per_node) \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn false \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/qwen_vl_chat/qlora/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-vl-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.7 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \