llama_prepare_model_devices: using device CUDA0 (Synthetic GPU) (synthetic-device) - 8192 MiB free
load_tensors: offloaded 20/33 layers to GPU
load_tensors:   CPU_Mapped model buffer size = 1152.00 MiB
load_tensors:        CUDA0 model buffer size = 4096.00 MiB
common_speculative_init_result: loading draft model 'synthetic-draft.gguf'
llama_prepare_model_devices: using device CUDA0 (Synthetic GPU) (synthetic-device) - 8192 MiB free
load_tensors: offloaded 7/7 layers to GPU
load_tensors:        CUDA0 model buffer size = 512.00 MiB
