ChatGLM-6B
ChatGLM-6B copied to clipboard
Deepspeed finetune需要多大的显存?
Is there an existing issue for this?
- [X] I have searched the existing issues
Current Behavior
3060 12G显卡deepspeed训练时报显存不够
Expected Behavior
No response
Steps To Reproduce
deepspeed训练需要多大显存?
Environment
OS: Ubuntu 22.04
Python: 3.10
Transformers: 4.26.1
PyTorch: 1.12
CUDA Support: True
Anything else?
No response
24GB batch size为1 都不够
测试 deepspeed zero2 62G /单张
看你训练样本,调整一个参数,再搞个量化也许能成,不过12G确实有点少
请问 显卡吞吐量如何 大概多少tokens/GPU/s
看你显卡型号
Duplicate of #556