# Captured from llama.cpp b10375 macOS arm64 llama-server, 2026-09-29.
# Model: ggml-org/gemma-4-12B-it-GGUF, Q8_0; --n-gpu-layers all, --ctx-size 512, --parallel 1.
# Startup lines selected from a successful /props-ready load; relative timestamps retained.
# Paths, endpoint port, and unrelated diagnostic text were omitted. Buffer figures are unmodified.
0.07.666.973 I llama_prepare_model_devices: using device MTL0 (Apple M5 Max) (unknown id) - 38338 MiB free
0.07.920.269 I load_tensors: offloaded 49/49 layers to GPU
0.07.920.270 I load_tensors:          CPU model buffer size =     0.00 MiB
0.07.920.270 I load_tensors:         MTL0 model buffer size =     0.00 MiB
0.07.920.980 I llama_context:        CPU  output buffer size =     1.00 MiB
0.07.921.044 I llama_kv_cache:       MTL0 KV buffer size =     0.00 MiB
0.07.921.952 I llama_kv_cache:       MTL0 KV buffer size =     0.00 MiB
0.07.930.858 I sched_reserve:       MTL0 compute buffer size =   106.02 MiB
0.07.930.860 I sched_reserve:        CPU compute buffer size =    16.02 MiB
0.07.930.860 I sched_reserve:        CPU compute buffer size =    16.02 MiB
0.07.973.012 I common_fit_params: fitting params to free memory took 0.43 seconds
0.08.094.022 I llama_prepare_model_devices: using device MTL0 (Apple M5 Max) (unknown id) - 38338 MiB free
0.10.603.053 I load_tensors: offloaded 49/49 layers to GPU
0.10.603.055 I load_tensors:   CPU_Mapped model buffer size =  1020.00 MiB
0.10.603.056 I load_tensors:  MTL0_Mapped model buffer size = 12067.63 MiB
0.10.607.475 I llama_context:        CPU  output buffer size =     1.00 MiB
0.10.607.728 I llama_kv_cache:       MTL0 KV buffer size =     8.00 MiB
0.10.621.283 I llama_kv_cache:       MTL0 KV buffer size =   160.00 MiB
0.10.638.697 I sched_reserve:       MTL0 compute buffer size =   106.02 MiB
0.10.638.701 I sched_reserve:        CPU compute buffer size =    16.02 MiB
