I encountered the error below mentioned while attempting to merge the adapter with the base model. I have attached my code for fine-tuning the model. Any assistance in resolving the issue would be greatly appreciated

notebook link.

RuntimeError: unable to mmap 9976576152 bytes from file </home/devuser/.cache/huggingface/hub/models–meta-llama–Llama-2-7b-chat-hf/snapshots/c1b0db933684edbfe29a06fa47eb19cc48025e93/model-00001-of-00002.safetensors>: Cannot allocate memory (12)

1 Like

Were you able to solve it? I am facing the same issue

Yep, I performed the merge adapter part in a separate file and I loaded the model with low_cpu usage.

1 Like

But wouldn’t this offload the entire model on the CPU, even if you have a GPU?

I don’t think so. Model offload scenarios can be handled by using the ‘device_map’ argument.