THUDM / GLM-4

GLM-4 series: Open Multilingual Multimodal Chat LMs | 开源多语言多模态对话模型
Apache License 2.0
5.28k stars 435 forks source link

lora微调和显卡兼容问题 #487

Closed znxd-wh closed 3 months ago

znxd-wh commented 3 months ago

System Info / 系統信息

cuda 12.2 transformers 4.43.3 python 3.10.12 操作系统 Ubuntu 22.04 cpu 架构x86_64(32核128线程) 内存 512G 显卡 NVIDIA GeForce RTX 4090-24G*8

Who can help? / 谁可以帮助到您?

No response

Information / 问题信息

Reproduction / 复现过程

CUDA_VISIBLE_DEVICES=4 python finetune.py AdvertiseGen/ /data/guanwei/LLM/glm-4-9b-chat configs/lora.yaml 进行lora微调,报错,感觉是和显卡兼容性问题 NotImplementedError: Using RTX 4000 series doesn't support faster communication broadband via P2P or IB. Please set NCCL_P2P_DISABLE="1" and NCCL_IB_DISABLE="1" or useaccelerate launch` which will do this automatically. image

Expected behavior / 期待表现

调试成功,保存checkpoint

zRzRzRzRzRzRzR commented 3 months ago

这个卡已经不支持了,这个模型本身也得用BF16调,FP16调不了

Lzzzii10 commented 2 months ago

@zRzRzRzRzRzRzR 也就是说这个卡不能做微调吗