Closed JianbangZ closed 2 days ago
我基于V2.5进行全量微调,只微调connector部分,冻结vision & LLM, 一百万张图文数据一个epoch训练时间为5-6小时,这个正常吗?我的机器是8XH100, 已经用了zero1 & flasth attention 2
No response
fintune_ds.sh
- OS: - Python: - Transformers: - PyTorch: - CUDA (`python -c 'import torch; print(torch.version.cuda)'`):
你好,我感觉这个速度应该是正常的,比我zero2和zero3的速度已经快太多了
是否已有关于该错误的issue或讨论? | Is there an existing issue / discussion for this?
该问题是否在FAQ中有解答? | Is there an existing answer for this in FAQ?
当前行为 | Current Behavior
我基于V2.5进行全量微调,只微调connector部分,冻结vision & LLM, 一百万张图文数据一个epoch训练时间为5-6小时,这个正常吗?我的机器是8XH100, 已经用了zero1 & flasth attention 2
期望行为 | Expected Behavior
No response
复现方法 | Steps To Reproduce
fintune_ds.sh
运行环境 | Environment
备注 | Anything else?
No response