Initial commit

swift-main/examples/pytorch/llm/scripts/codefuse_codellama_34b/lora/sft.sh

+# Experimental environment: V100, A10, 3090
+# 18GB GPU memory
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_sft.py \
+    --model_type codefuse-codellama-34b-chat \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype fp16 \
+    --output_dir output \
+    --dataset xxx.jsonl \
+    --val_dataset yyy.jsonl \
+    --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 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 true \

swift-main/examples/pytorch/llm/scripts/codegeex2_6b/lora_ddp_ds/infer.sh

+# Experimental environment: 3090
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/codegeex2-6b/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/codegeex2_6b/lora_ddp_ds/sft.sh

+# Experimental environment: 2 * 3090
+# 2 * 20GB 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 codegeex2-6b \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset leetcode-python-en \
+    --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 \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/codeqwen1half_7b_chat/lora/infer.sh

+# Experimental environment: 3090
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/codeqwen1half-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/codeqwen1half_7b_chat/lora/sft.sh

+# Experimental environment: 3090，A10，V100...
+# 20GB GPU memory
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+    --model_type codeqwen1half-7b-chat \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset leetcode-python-en \
+    --num_train_epochs 3 \
+    --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 \

swift-main/examples/pytorch/llm/scripts/codeqwen1half_7b_chat_awq/lora/infer.sh

+# Experiment env: A10, RTX3090/4090, A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/codeqwen1half-7b-chat-awq/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 \
+    --stream false \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/codeqwen1half_7b_chat_awq/lora/sft.sh

+# Experiment env: A10, RTX3090/4090, A100
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+  --model_type codeqwen1half-7b-chat-awq \
+  --dataset leetcode-python-en \
+  --batch_size 4 \
+  --max_length 2048 \
+  --gradient_accumulation_steps 2 \
+  --learning_rate 5e-5 \
+  --use_flash_attn true \
+  --eval_steps 2000 \
+  --save_steps 2000 \
+  --num_train_epochs 3 \
+  --check_dataset_strategy none \
+  --gradient_checkpointing true \
+  --weight_decay 0.1 \
+  --max_grad_norm 1.0 \
+  --warmup_ratio 0.03 \
+  --save_total_limit 2 \
+  --logging_steps 10 \
+  --sft_type lora \
+  --lora_target_modules ALL \
+  --lora_rank 8 \
+  --lora_alpha 32

swift-main/examples/pytorch/llm/scripts/cogagent_18b_chat/lora/infer.sh

+# Experimental environment: V100, A10, 3090
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/cogagent-18b-chat/vx-xxx/checkpoint-xx" \
+    --load_args_from_ckpt_dir true \
+    --eval_human true \
+    --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/cogagent_18b_chat/lora/sft.sh

+# Experimental environment: 2 * A100
+# 2 * 45GB
+CUDA_VISIBLE_DEVICES=0,1 \
+swift sft \
+    --model_type cogagent-18b-chat \
+    --sft_type lora \
+    --tuner_backend peft \
+    --dtype AUTO \
+    --output_dir output \
+    --dataset coco-en-2-mini \
+    --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 DEFAULT \
+    --gradient_checkpointing false \
+    --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/custom/tigerbot_13b_chat/qlora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/tigerbot-13b-chat/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --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/custom/tigerbot_13b_chat/qlora_ddp_ds/sft.sh

+# Experimental environment: 2 * 3090
+# 2 * 12GB 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 tigerbot-13b-chat \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset stsb-en \
+    --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 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 \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/custom/tigerbot_7b/lora_ddp_ds/infer.sh

+# Experimental environment: A10
+PYTHONPATH=../../.. \
+CUDA_VISIBLE_DEVICES=0 \
+python llm_infer.py \
+    --ckpt_dir "output/tigerbot-13b/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --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/custom/tigerbot_7b/lora_ddp_ds/sft.sh

+# Experimental environment: 2 * 3090
+# 2 * 16GB 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 tigerbot-7b \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type default-generation \
+    --dtype AUTO \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset stsb-en \
+    --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 \
+    --deepspeed default-zero2 \

swift-main/examples/pytorch/llm/scripts/dbrx-instruct/lora_mp/infer.sh

+# Experimental environment: 4 * A100
+# 4 * 65GB GPU memory
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+swift infer \
+    --ckpt_dir "output/dbrx-instruct/vx-xxx/checkpoint-xxx" \
+    --load_dataset_config true \
+    --use_flash_attn true \
+    --temperature 0.3 \
+    --top_p 0.7 \
+    --repetition_penalty 1. \
+    --do_sample true \
+    --merge_lora false \

swift-main/examples/pytorch/llm/scripts/dbrx-instruct/lora_mp/sft.sh

+# Experimental environment: 4 * A100
+# 4 * 74GB GPU memory
+CUDA_VISIBLE_DEVICES=0,1,2,3 \
+swift sft \
+    --model_type dbrx-instruct \
+    --model_revision master \
+    --sft_type lora \
+    --tuner_backend peft \
+    --template_type AUTO \
+    --dtype bf16 \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset blossom-math-zh \
+    --num_train_epochs 1 \
+    --max_length 1024 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_target_modules ALL \
+    --lora_dtype AUTO \
+    --gradient_checkpointing false \
+    --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 true

swift-main/examples/pytorch/llm/scripts/deepseek-v2-chat/lora_mp/deepseek2_device_map.json

+{
+    "model.embed_tokens": "cuda:0",
+    "model.layers.0": "cuda:0",
+    "model.layers.1": "cuda:0",
+    "model.layers.2": "cuda:0",
+    "model.layers.3": "cuda:0",
+    "model.layers.4": "cuda:0",
+    "model.layers.5": "cuda:0",
+    "model.layers.6": "cuda:0",
+    "model.layers.7": "cuda:1",
+    "model.layers.8": "cuda:1",
+    "model.layers.9": "cuda:1",
+    "model.layers.10": "cuda:1",
+    "model.layers.11": "cuda:1",
+    "model.layers.12": "cuda:1",
+    "model.layers.13": "cuda:1",
+    "model.layers.14": "cuda:2",
+    "model.layers.15": "cuda:2",
+    "model.layers.16": "cuda:2",
+    "model.layers.17": "cuda:2",
+    "model.layers.18": "cuda:2",
+    "model.layers.19": "cuda:2",
+    "model.layers.20": "cuda:2",
+    "model.layers.21": "cuda:3",
+    "model.layers.22": "cuda:3",
+    "model.layers.23": "cuda:3",
+    "model.layers.24": "cuda:3",
+    "model.layers.25": "cuda:3",
+    "model.layers.26": "cuda:3",
+    "model.layers.27": "cuda:3",
+    "model.layers.28": "cuda:4",
+    "model.layers.29": "cuda:4",
+    "model.layers.30": "cuda:4",
+    "model.layers.31": "cuda:4",
+    "model.layers.32": "cuda:4",
+    "model.layers.33": "cuda:4",
+    "model.layers.34": "cuda:4",
+    "model.layers.35": "cuda:4",
+    "model.layers.36": "cuda:5",
+    "model.layers.37": "cuda:5",
+    "model.layers.38": "cuda:5",
+    "model.layers.39": "cuda:5",
+    "model.layers.40": "cuda:5",
+    "model.layers.41": "cuda:5",
+    "model.layers.42": "cuda:5",
+    "model.layers.43": "cuda:5",
+    "model.layers.44": "cuda:6",
+    "model.layers.45": "cuda:6",
+    "model.layers.46": "cuda:6",
+    "model.layers.47": "cuda:6",
+    "model.layers.48": "cuda:6",
+    "model.layers.49": "cuda:6",
+    "model.layers.50": "cuda:6",
+    "model.layers.51": "cuda:6",
+    "model.layers.52": "cuda:7",
+    "model.layers.53": "cuda:7",
+    "model.layers.54": "cuda:7",
+    "model.layers.55": "cuda:7",
+    "model.layers.56": "cuda:7",
+    "model.layers.57": "cuda:7",
+    "model.layers.58": "cuda:7",
+    "model.layers.59": "cuda:7",
+    "model.norm": "cuda:7",
+    "lm_head": "cuda:7"
+}

swift-main/examples/pytorch/llm/scripts/deepseek-v2-chat/lora_mp/infer.sh

+# Experimental environment: 8*A100
+# cd /path/to/swift/example/pytorch/llm
+
+PYTHONPATH=../../.. \
+
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+python llm_infer.py \
+    --ckpt_dir output/deepseek-v2-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 \
+    --device_map_config_path scripts/deepseek-v2-chat/lora_ddp_ds3/deepseek2_device_map.json

swift-main/examples/pytorch/llm/scripts/deepseek-v2-chat/lora_mp/sft.sh

+# Experimental environment: 8*A100
+# 8*80GB GPU memory
+CUDA_VISIBLE_DEVICES=0,1,2,3,4,5,6,7 \
+swift sft \
+    --model_type deepseek-v2-chat \
+    --sft_type lora \
+    --tuner_backend peft \
+    --dtype bf16 \
+    --output_dir output \
+    --ddp_backend nccl \
+    --dataset alpaca-zh#5000 \
+    --num_train_epochs 1 \
+    --max_length 1024 \
+    --check_dataset_strategy warning \
+    --lora_rank 8 \
+    --lora_alpha 32 \
+    --lora_dropout_p 0.05 \
+    --lora_dtype AUTO \
+    --lora_target_modules DEFAULT \
+    --gradient_checkpointing false \
+    --use_flash_attn 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 10 \
+    --logging_steps 10 \
+    --device_map_config_path scripts/deepseek-v2-chat/lora_ddp_ds3/deepseek2_device_map.json

swift-main/examples/pytorch/llm/scripts/deepseek_moe_16b_chat/lora/infer.sh

+# Experimental environment: A100
+CUDA_VISIBLE_DEVICES=0 \
+swift infer \
+    --ckpt_dir "output/deepseek-moe-16b-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/deepseek_moe_16b_chat/lora/sft.sh

+# Experimental environment: A100
+# 52GB GPU memory
+CUDA_VISIBLE_DEVICES=0 \
+swift sft \
+    --model_type deepseek-moe-16b-chat \
+    --dataset damo-agent-mini-zh \
+    --train_dataset_sample 20000 \
+    --max_length 4096 \
+    --gradient_checkpointing true \
+    --eval_steps 100 \
+    --use_flash_attn true \
+    --output_dir output \