Loading checkpoint shards: 0% Killed
#3
by
yueyueyushi
- opened
Every time I load the codegen25-7b-mono locally, whether through checkpoint or directly downloading the corresponding files required for codegen25 model weights and inference online in the huggingface, I encounter issues with the title. I confirm that I have installed all the required dependencies and can run codegen-350m-mono locally using a V100, 32G memory GPU
40G memory is not enough. 10G Virtual memory needs to be set to solve the problem of loading shards 0%&killed. But I encountered a new problem: during the inference process, the GPU was not used at all, and there were no programs executing in the NVidia-smi process