deepseek-ai / deepseek-ai/DeepSeek-OCR
12G 显存跑推理时爆显存（已修复）
Open
- Dominant language
- Python
- Stars
- 23.9k
- Forks
- 2.2k
- PR merge metrics
- No merged PRs in 30d
Description
我用3060（12G）跑README.md中下面这段代码是爆显存了：
cd DeepSeek-OCR-master/DeepSeek-OCR-hf
python run_dpsk_ocr.py
原因在于代码中
model = model.eval().cuda().to(torch.bfloat16)
这一行，在 .cuda() 的时候已经爆了，改成下面这样先变 bf16 再.cuda() 就没问题了
model = model.eval().to(torch.bfloat16).cuda()
Contributor guide
No contributing guide indexed for this repository
Assessment
This issue has not been assessed yet.