Initial commit

swift-main/examples/pytorch/llm/scripts/openbuddy_llama2_13b_chat/qlora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/openbuddy-llama2-13b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config 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/openbuddy_llama2_13b_chat/qlora_ddp_ds/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 11GB 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 OpenBuddy/openbuddy-llama2-13b-v8.1-fp16 \
+    --model_revision master \
+    --sft_type lora \
+    --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 1 \
+    --max_length 2048 \
+    --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 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 \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/openbuddy_llama2_70b_chat/qlora_ddp_ds/infer.sh

+# Experimental environment: A100
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/openbuddy-llama2-70b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config 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/openbuddy_llama2_70b_chat/qlora_ddp_ds/sft.sh

+# Experimental environment: 2 * A100
+# 2 * 40GB 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 OpenBuddy/openbuddy-llama2-70b-v10.1-bf16 \
+    --model_revision master \
+    --sft_type lora \
+    --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 1 \
+    --max_length 2048 \
+    --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 q_proj v_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 \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/openbuddy_llama2_70b_chat/qlora_mp/infer.sh

+# Experimental environment: 2 * 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1 \
+python llm_infer.py \
+    --ckpt_dir "output/openbuddy-llama2-70b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config 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/openbuddy_llama2_70b_chat/qlora_mp/sft.sh

+# Experimental environment: 2 * 3090
+# 2 * 23GB GPU memory
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0,1 \
+python llm_sft.py \
+    --model_id_or_path OpenBuddy/openbuddy-llama2-70b-v10.1-bf16 \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --dataset blossom-math-zh \
+    --train_dataset_sample -1 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --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 q_proj v_proj \
+    --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 \

swift-main/examples/pytorch/llm/scripts/openbuddy_mistral_7b_chat/lora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/openbuddy-mistral-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config 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/openbuddy_mistral_7b_chat/lora_ddp_ds/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 18GB 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 OpenBuddy/openbuddy-mistral-7b-v17.1-32k \
+    --model_revision master \
+    --sft_type lora \
+    --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 1 \
+    --max_length 2048 \
+    --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 \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/openbuddy_mistral_7b_chat/lora_mp_ddp/infer.sh

+# Experimental environment: 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/openbuddy-mistral-7b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config 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/openbuddy_mistral_7b_chat/lora_mp_ddp/sft.sh

+# Experimental environment: 4 * 3090
+# 4 * 19GB GPU memory
+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 OpenBuddy/openbuddy-mistral-7b-v17.1-32k \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --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 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 \

swift-main/examples/pytorch/llm/scripts/orion_14b_chat/lora_ddp/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/orion-14b-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/orion_14b_chat/lora_ddp/sft.sh

+# Experimental environment: 4 * A100
+# 4 * 45GB GPU memory
+nproc_per_node=4
+
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+NPROC_PER_NODE=$nproc_per_node \
+MASTER_PORT=29500 \
+swift sft \
+    --model_type orion-14b-chat \
+    --sft_type lora \
+    --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 2 \
+    --max_length 2048 \
+    --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 true \

swift-main/examples/pytorch/llm/scripts/phi2_3b/lora/infer.sh

+# Experimental environment: A10
+# 8GB GPU memory
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/phi2-3b/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. \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/phi2_3b/lora/sft.sh

+# Experimental environment: A100
+# 60GB GPU memory
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+    --model_type phi2-3b \
+    --sft_type lora \
+    --template_type default \
+    --train_dataset_sample 20000 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --learning_rate 1e-4 \
+    --use_flash_attn true \
+    --lora_target_modules ALL \
+    --dataset codefuse-python-en \
+    --gradient_checkpointing false \

swift-main/examples/pytorch/llm/scripts/phi3_4b_4k_instruct/lora/infer.sh

+# Experimental environment: A10
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/phi3-4b-4k-instruct/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. \

swift-main/examples/pytorch/llm/scripts/phi3_4b_4k_instruct/lora/sft.sh

+# Experimental environment: A10
+# 12GB GPU memory
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+    --model_type phi3-4b-4k-instruct \
+    --sft_type lora \
+    --template_type AUTO \
+    --train_dataset_sample 20000 \
+    --eval_steps 100 \
+    --output_dir output \
+    --num_train_epochs 3 \
+    --max_length 4096 \
+    --learning_rate 1e-4 \
+    --use_flash_attn false \
+    --lora_target_modules ALL \
+    --dataset codefuse-python-en \

swift-main/examples/pytorch/llm/scripts/polylm_13b/qlora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/polylm-13b/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --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/polylm_13b/qlora_ddp_ds/sft.sh

+# Experimental environment: 2 * A10
+# 2 * 13GB 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 damo/nlp_polylm_13b_text_generation \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type default-generation \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset advertise-gen-zh \
+    --train_dataset_sample 20000 \
+    --num_train_epochs 1 \
+    --max_length 2048 \
+    --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 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 \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/qwen1half_110b_chat/lora_ddp_ds/infer.sh

+# Experimental environment: 4*A100
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+swift infer \
+    --ckpt_dir "output/qwen1half-110b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \
+    --max_new_tokens 2048 \
+    --temperature 0.5 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/qwen1half_110b_chat/lora_ddp_ds/sft.sh

+# Experimental environment: 4*A100
+# 4*75GB GPU memory
+nproc_per_node=4
+CUDA_VISIBLE_DEVICES=0,1,2,3\
+NPROC_PER_NODE=$nproc_per_node \
+swift sft \
+    --model_type qwen1half-110b-chat \
+    --sft_type lora \
+    --tuner_backend peft \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset alpaca-zh \
+    --train_dataset_sample -1 \
+    --num_train_epochs 2 \
+    --max_length 2048 \
+    --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 true \
+    --deepspeed default-zero3