AI开发平台MODELARTS-InternVL2基于DevServer适配PyTorch NPU训练指导（6.3.909）:Step9 开始训练

时间：2024-10-24 17:19:46

AI开发平台MODELARTS AIGC模型训练推理

Step9 开始训练

单机训练

cd ${container_work_dir}/InternVL/internvl_chat 
# 8B全参微调
GPUS=8 PER_DEVICE_BATCH_SIZE=2 sh shell/internvl2.0/2nd_finetune/internvl2_8b_internlm2_7b_dynamic_res_2nd_finetune_full.sh 
# 8Blora微调
GPUS=8 PER_DEVICE_BATCH_SIZE=2 sh shell/internvl2.0/2nd_finetune/internvl2_8b_internlm2_7b_dynamic_res_2nd_finetune_lora.sh 
# 26Blora微调
GPUS=8 PER_DEVICE_BATCH_SIZE=2 sh shell/internvl2.0/2nd_finetune/internvl2_26b_internlm2_20b_dynamic_res_2nd_finetune_lora.sh 
# 40Blora微调
GPUS=8 PER_DEVICE_BATCH_SIZE=2 sh shell/internvl2.0/2nd_finetune/internvl2_40b_hermes2_yi_34b_dynamic_res_2nd_finetune_lora.sh

多机训练

cd ${container_work_dir}/InternVL/internvl_chat  
# 8B lora
GPUS=8 PER_DEVICE_BATCH_SIZE=2 NNODES=${NODE_NUM} NODE_RANK=${NODE_RANK} MASTER_ADDR="${master_node_ip}" sh shell/internvl2.0/2nd_finetune/internvl2_8b_internlm2_7b_dynamic_res_2nd_finetune_lora_multi.sh 
# 8B full
GPUS=8 PER_DEVICE_BATCH_SIZE=2 NNODES=${NODE_NUM} NODE_RANK=${NODE_RANK} MASTER_ADDR="${master_node_ip}" sh shell/internvl2.0/2nd_finetune/internvl2_8b_internlm2_7b_dynamic_res_2nd_finetune_full_multi.sh 
# 26B lora
GPUS=8 PER_DEVICE_BATCH_SIZE=2 NNODES=${NODE_NUM} NODE_RANK=${NODE_RANK} MASTER_ADDR="${master_node_ip}" sh shell/internvl2.0/2nd_finetune/internvl2_26b_internlm2_20b_dynamic_res_2nd_finetune_lora_multi.sh 
# 26B full
GPUS=8 PER_DEVICE_BATCH_SIZE=2 NNODES=${NODE_NUM} NODE_RANK=${NODE_RANK} MASTER_ADDR="${master_node_ip}" sh shell/internvl2.0/2nd_finetune/internvl2_26b_internlm2_20b_dynamic_res_2nd_finetune_full_multi.sh

参数说明：