01-ai

/

Yi-34B

like 1.26k

Some modules are dispatched on the CPU or the disk. Make sure you have enough GPU RAM to fit

                the quantized model. If you want to dispatch the model on the CPU or the disk while keeping

                these modules in 32-bit, you need to set `load_in_8bit_fp32_cpu_offload=True` and pass a custom

                `device_map` to `from_pretrained`. Check

                https://huggingface.co/docs/transformers/main/en/main_classes/quantization#offload-between-cpu-and-gpu

                for more details.