$ /Users/minipro/quantdiff/llama.cpp/llama-b11770/llama-bench -m /Users/minipro/quantdiff/models/campaign/d54834c785c7/LFM2.5-24B-A2B-UD-Q3_K_XL.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json # started 2026-10-25T05:58:46Z ggml_metal_device_init: tensor API disabled for pre-M5 and pre-A19 devices ggml_metal_library_init: using embedded metal library ggml_metal_library_compile_all: compiled 'fa_aux' library in 0.001 sec ggml_metal_library_compile_all: compiled 'fa_f16' library in 0.002 sec ggml_metal_library_compile_all: compiled 'fa_f32' library in 0.002 sec ggml_metal_library_compile_all: compiled 'fa_q4_0' library in 0.002 sec ggml_metal_library_compile_all: compiled 'fa_q4_1' library in 0.008 sec ggml_metal_library_compile_all: compiled 'fa_q5_0' library in 0.003 sec ggml_metal_library_compile_all: compiled 'fa_q5_1' library in 0.013 sec ggml_metal_library_compile_all: compiled 'fa_q8_0' library in 0.012 sec ggml_metal_library_compile_all: compiled 'fa_tensor' library in 0.008 sec ggml_metal_library_compile_all: compiled 'fa_vec_f16' library in 0.007 sec ggml_metal_library_compile_all: compiled 'fa_vec_f32' library in 0.010 sec ggml_metal_library_compile_all: compiled 'fa_vec_q4_0' library in 0.006 sec ggml_metal_library_compile_all: compiled 'fa_vec_q4_1' library in 0.010 sec ggml_metal_library_compile_all: compiled 'fa_vec_q5_0' library in 0.012 sec ggml_metal_library_compile_all: compiled 'fa_vec_q5_1' library in 0.009 sec ggml_metal_library_compile_all: compiled 'fa_vec_q8_0' library in 0.005 sec ggml_metal_library_compile_all: compiled 'mul_mv' library in 0.004 sec ggml_metal_library_compile_all: compiled 'mul_mv_mma' library in 0.015 sec ggml_metal_library_compile_all: compiled 'mul_mm' library in 0.011 sec ggml_metal_library_compile_all: compiled 'quantize' library in 0.013 sec ggml_metal_library_compile_all: compiled 'softmax' library in 0.002 sec ggml_metal_library_compile_all: compiled 'norm' library in 0.002 sec ggml_metal_library_compile_all: compiled 'unary' library in 0.002 sec ggml_metal_library_compile_all: compiled 'binbcast' library in 0.013 sec ggml_metal_library_compile_all: compiled 'reduce' library in 0.013 sec ggml_metal_library_compile_all: compiled 'tri' library in 0.013 sec ggml_metal_library_compile_all: compiled 'ssm' library in 0.011 sec ggml_metal_library_compile_all: compiled 'wkv' library in 0.011 sec ggml_metal_library_compile_all: compiled 'gated_delta_net' library in 0.007 sec ggml_metal_library_compile_all: compiled 'solve_tri' library in 0.007 sec ggml_metal_library_compile_all: compiled 'rope' library in 0.007 sec ggml_metal_library_compile_all: compiled 'conv' library in 0.013 sec ggml_metal_library_compile_all: compiled 'upscale' library in 0.007 sec ggml_metal_library_compile_all: compiled 'argsort' library in 0.013 sec ggml_metal_library_compile_all: compiled 'pool' library in 0.003 sec ggml_metal_library_compile_all: compiled 'misc' library in 0.003 sec ggml_metal_library_compile_all: loaded 36 libraries from embedded data in 0.015 sec (max single = 0.015 sec) ggml_metal_rsets_init: creating a residency set collection (keep_alive = 180 s) ggml_metal_device_init: GPU name: MTL0 (Apple M2 Pro) ggml_metal_device_init: GPU family: MTLGPUFamilyApple8 (1008) ggml_metal_device_init: GPU family: MTLGPUFamilyMetal4 (5002) ggml_metal_device_init: simdgroup reduction = true ggml_metal_device_init: simdgroup matrix mul. = true ggml_metal_device_init: has unified memory = true ggml_metal_device_init: has bfloat = true ggml_metal_device_init: has tensor = false ggml_metal_device_init: use residency sets = true ggml_metal_device_init: use shared buffers = true ggml_metal_device_init: recommendedMaxWorkingSetSize = 27917.29 MB [ { "build_commit": "39da130fc", "build_number": 11770, "cpu_info": "Accelerate, Apple M2 Pro", "gpu_info": "Apple M2 Pro", "backends": "MTL,BLAS", "model_filename": "/Users/minipro/quantdiff/models/campaign/d54834c785c7/LFM2.5-24B-A2B-UD-Q3_K_XL.gguf", "model_type": "lfm2moe 24B.A2B Q4_K - Small", "model_size": 10938849792, "model_n_params": 23971587712, "n_batch": 2048, "n_ubatch": 512, "n_threads": 8, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": 1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 512, "n_gen": 0, "n_depth": 0, "test_time": "2026-10-25T05:58:47Z", "avg_ns": 641653200, "stddev_ns": 1657713, "avg_ts": 797.943084, "stddev_ts": 2.061003, "samples_ns": [ 640687250, 643655417, 639656250, 643028208, 641238875 ], "samples_ts": [ 799.142, 795.457, 800.43, 796.233, 798.454 ] }, { "build_commit": "39da130fc", "build_number": 11770, "cpu_info": "Accelerate, Apple M2 Pro", "gpu_info": "Apple M2 Pro", "backends": "MTL,BLAS", "model_filename": "/Users/minipro/quantdiff/models/campaign/d54834c785c7/LFM2.5-24B-A2B-UD-Q3_K_XL.gguf", "model_type": "lfm2moe 24B.A2B Q4_K - Small", "model_size": 10938849792, "model_n_params": 23971587712, "n_batch": 2048, "n_ubatch": 512, "n_threads": 8, "cpu_mask": "0x0", "cpu_strict": false, "poll": 50, "type_k": "f16", "type_v": "f16", "n_gpu_layers": 999, "n_cpu_moe": 0, "split_mode": "layer", "main_gpu": 0, "no_kv_offload": false, "flash_attn": 1, "devices": "auto", "tensor_split": "0.00", "tensor_buft_overrides": "none", "load_mode": "auto", "lazy_mode": "auto", "embeddings": false, "no_op_offload": 0, "no_host": false, "repack": true, "fit_target": 0, "fit_min_ctx": 0, "n_prompt": 0, "n_gen": 128, "n_depth": 0, "test_time": "2026-10-25T05:58:51Z", "avg_ns": 1858984050, "stddev_ns": 2360266, "avg_ts": 68.854902, "stddev_ts": 0.087462, "samples_ns": [ 1855441417, 1858717875, 1858782291, 1860100708, 1861877959 ], "samples_ts": [ 68.9863, 68.8647, 68.8623, 68.8135, 68.7478 ] } ] # exit 0 after 14s