Initial commit

swift-main/examples/pytorch/llm/scripts/qwen_7b/lora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-7b/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_7b/lora_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 \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type default-generation \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset dureader-robust-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 c_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_7b/qlora_ddp/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-7b/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_7b/qlora_ddp/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 14GB 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 \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type default-generation \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset tigerbot-law-zh \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 4096 \
+    --check_dataset_strategy warning \
+    --quantization_bit 4 \
+    --bnb_4bit_comp_dtype AUTO \
+    --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 \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full/sft.sh

+# Experimental environment: A100
+# 80GB GPU memory
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+    --model_type qwen-7b-chat \
+    --sft_type full \
+    --train_dataset_sample -1 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 1 \
+    --max_length 4096 \
+    --learning_rate 1e-5 \
+    --use_flash_attn true \
+    --save_only_model true \
+    --dataset codefuse-evol-instruction-zh \
+    --preprocess_num_proc 4 \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_ddp_zero2/infer.sh

+# Experimental environment: 3090
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \
+    --max_new_tokens 2048 \
+    --temperature 0.1 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_ddp_zero2/sft.sh

+# Experimental environment: 4 * A100
+# 4 * 68GB GPU memory
+nproc_per_node=4
+
+NPROC_PER_NODE=$nproc_per_node \
+MASTER_PORT=29500 \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+swift sft \
+    --model_id_or_path qwen/Qwen-7B-Chat \
+    --model_revision master \
+    --sft_type full \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset blossom-math-zh \
+    --train_dataset_sample -1 \
+    --num_train_epochs 5 \
+    --max_length 2048 \
+    --check_dataset_strategy warning \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps $(expr 64 / $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 true \
+    --deepspeed 'default-zero2' \
+    --save_only_model true \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_ddp_zero3/infer.sh

+# Experimental environment: 3090
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \
+    --max_new_tokens 2048 \
+    --temperature 0.1 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_ddp_zero3/sft.sh

+# Experimental environment: 4 * A100
+# 4 * 55GB GPU memory
+nproc_per_node=4
+
+NPROC_PER_NODE=$nproc_per_node \
+MASTER_PORT=29500 \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+swift sft \
+    --model_id_or_path qwen/Qwen-7B-Chat \
+    --model_revision master \
+    --sft_type full \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset blossom-math-zh \
+    --train_dataset_sample -1 \
+    --num_train_epochs 5 \
+    --max_length 2048 \
+    --check_dataset_strategy warning \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-4 \
+    --gradient_accumulation_steps $(expr 64 / $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 true \
+    --deepspeed 'default-zero3' \
+    --save_only_model true \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_freeze_ddp/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_freeze_ddp/sft.sh

+# Experimental environment: 2 * A100
+# 2 * 80GB GPU memory
+NPROC_PER_NODE=2 \
+CUDA_VISIBLE_DEVICES=0,1 \
+swift sft \
+    --model_type qwen-7b-chat \
+    --sft_type full \
+    --train_dataset_sample -1 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 1 \
+    --max_length 4096 \
+    --learning_rate 1e-5 \
+    --use_flash_attn true \
+    --save_only_model true \
+    --dataset codefuse-evol-instruction-zh \
+    --freeze_parameters 0.25 \
+    --additional_trainable_parameters transformer.wte \
+    --preprocess_num_proc 4 \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_mp/infer.sh

+# Experimental environment: A10, 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \
+    --max_new_tokens 2048 \
+    --temperature 0.1 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_mp/sft.sh

+# Experimental environment: 2 * A100
+# 2 * 55GB GPU memory (use flash_attn)
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1 \
+python llm_sft.py \
+    --model_id_or_path qwen/Qwen-7B-Chat \
+    --model_revision master \
+    --sft_type full \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --dataset damo-agent-zh \
+    --train_dataset_sample 200000 \
+    --num_train_epochs 1 \
+    --max_length 8192 \
+    --check_dataset_strategy warning \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-5 \
+    --gradient_accumulation_steps 16 \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_only_model true \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn true \
+    --preprocess_num_proc 4 \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_mp_ddp/infer.sh

+# Experimental environment: A10, 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \
+    --max_new_tokens 2048 \
+    --temperature 0.7 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/full_mp_ddp/sft.sh

+# Experimental environment: 4 * A100
+# 4 * 55GB GPU memory (use flash_attn)
+nproc_per_node=2
+
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+torchrun \
+    --nproc_per_node=$nproc_per_node \
+    --master_port 29500 \
+    llm_sft.py \
+    --model_id_or_path qwen/Qwen-7B-Chat \
+    --model_revision master \
+    --sft_type full \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --dataset medical-en medical-zh \
+    --train_dataset_sample 200000 \
+    --num_train_epochs 1 \
+    --max_length 8192 \
+    --check_dataset_strategy warning \
+    --gradient_checkpointing true \
+    --batch_size 1 \
+    --weight_decay 0.1 \
+    --learning_rate 1e-5 \
+    --gradient_accumulation_steps $(expr 16 / $nproc_per_node) \
+    --max_grad_norm 0.5 \
+    --warmup_ratio 0.03 \
+    --eval_steps 100 \
+    --save_steps 100 \
+    --save_only_model true \
+    --save_total_limit 2 \
+    --logging_steps 10 \
+    --use_flash_attn true \
+    --preprocess_num_proc 4 \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/lora/eval.sh

+CUDA_VISIBLE_DEVICES=0 \
+swift eval \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --eval_dataset arc \
+    --eval_limit 10 \
+    --infer_backend pt \
+    --custom_eval_config eval_example/custom_config.json

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/lora/infer.sh

+# Experimental environment: V100, A10, 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.1 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/lora/sft.sh

+# Experimental environment: A100
+# 24GB GPU memory
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_sft.py \
+    --model_id_or_path qwen/Qwen-7B-Chat \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --dataset ms-agent \
+    --use_loss_scale true \
+    --train_dataset_mix_ratio 2.0 \
+    --train_dataset_sample -1 \
+    --num_train_epochs 2 \
+    --max_length 1500 \
+    --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 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 \
+    --self_cognition_sample 3000 \
+    --model_name 卡卡罗特 \
+    --model_author 陶白白 \

swift-main/examples/pytorch/llm/scripts/qwen_7b_chat/lora_ddp/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/qwen-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn false \
+    --max_new_tokens 2048 \
+    --temperature 0.1 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \