When I use the command āpython convert_llama_weights_to_hf.py --input_dir /home/hadoop-kg-llm-ddpt/llama/ --model_size 70B --output_dir /home/hadoop-kg-llm-ddpt/llama/Llama-2-70b-chat-hfā to convert llama2, an error occurred:
Also, I noticed
Important note: you need to be able to host the whole model in RAM to execute this script (even if the biggest versions
come in several checkpoints they each contain a part of each weight of the model, so we need to load them all in RAM).
However, I still want to ask, when the memory is not enough, is there any parameter control that can use the disk instead of the memory to complete the task?