Tongkun Su

icon company University of Chinese Academy of Sciences icon location Shenzhen, China icon location Master's student @UCAS @SIAT, medical report generation & medical VQA & medical multimodal pretraining

Results 3 issues of Tongkun Su

使用Lora微调加载参数时自动结束进程

37 comment

使用Lora微调时代码运行到加载预训练权重时会自动结束进程，也没有报错信息，请问有人知道是为什么吗？ 用QLora微调时可以正常运行。 下面是运行的Log (pytorch-gpu) [xxx@localhost VisualGLM-6B]$ sh finetune_visualglm.sh NCCL_DEBUG=info NCCL_IB_DISABLE=0 NCCL_NET_GDR_LEVEL=2 deepspeed --master_port 16666 --hostfile hostfile_single finetune_visualglm.py --experiment-name finetune-visualglm-6b --model-parallel-size 1 --mode finetune --train-iters 300 --resume-dataloader --max_source_length 64 --max_target_length 256...

模型训练次数

1 comment

请问微调了多少个iteration才达到demo中的效果

Language

2 comment

Hello, thank you very much for your work! I'd like to ask if there are any language restrictions on topicGPT. Can it handle non-English text?