$ /Users/minipro/quantdiff/work/llama-b11869/src/build/bin/llama-bench -m /Users/minipro/quantdiff/models/campaign/3376b755f9ab/Sage-A1.5.i1-IQ3_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json # started 2026-10-25T20:05:23Z 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 0.008 sec ggml_metal_library_compile_all: compiled 'fa_f16' library in 0.051 sec ggml_metal_library_compile_all: compiled 'fa_f32' library in 0.054 sec ggml_metal_library_compile_all: compiled 'fa_q4_0' library in 0.052 sec ggml_metal_library_compile_all: compiled 'fa_q4_1' library in 0.039 sec ggml_metal_library_compile_all: compiled 'fa_q5_0' library in 0.055 sec ggml_metal_library_compile_all: compiled 'fa_q5_1' library in 0.018 sec ggml_metal_library_compile_all: compiled 'fa_q8_0' library in 0.056 sec ggml_metal_library_compile_all: compiled 'fa_tensor' library in 0.006 sec ggml_metal_library_compile_all: compiled 'fa_vec_f16' library in 0.035 sec ggml_metal_library_compile_all: compiled 'fa_vec_f32' library in 0.054 sec ggml_metal_library_compile_all: compiled 'fa_vec_q4_0' library in 0.027 sec ggml_metal_library_compile_all: compiled 'fa_vec_q4_1' library in 0.049 sec ggml_metal_library_compile_all: compiled 'fa_vec_q5_0' library in 0.038 sec ggml_metal_library_compile_all: compiled 'fa_vec_q5_1' library in 0.017 sec ggml_metal_library_compile_all: compiled 'fa_vec_q8_0' library in 0.022 sec ggml_metal_library_compile_all: compiled 'mul_mv' library in 0.012 sec ggml_metal_library_compile_all: compiled 'mul_mv_mma' library in 0.031 sec ggml_metal_library_compile_all: compiled 'mul_mm' library in 0.045 sec ggml_metal_library_compile_all: compiled 'quantize' library in 0.041 sec ggml_metal_library_compile_all: compiled 'softmax' library in 0.006 sec ggml_metal_library_compile_all: compiled 'norm' library in 0.007 sec ggml_metal_library_compile_all: compiled 'unary' library in 0.006 sec ggml_metal_library_compile_all: compiled 'binbcast' library in 0.008 sec ggml_metal_library_compile_all: compiled 'reduce' library in 0.013 sec ggml_metal_library_compile_all: compiled 'tri' library in 0.045 sec ggml_metal_library_compile_all: compiled 'ssm' library in 0.006 sec ggml_metal_library_compile_all: compiled 'wkv' library in 0.007 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 0.039 sec ggml_metal_library_compile_all: compiled 'rope' library in 0.006 sec ggml_metal_library_compile_all: compiled 'conv' library in 0.055 sec ggml_metal_library_compile_all: compiled 'upscale' library in 0.020 sec ggml_metal_library_compile_all: compiled 'argsort' library in 0.007 sec ggml_metal_library_compile_all: compiled 'pool' library in 0.007 sec ggml_metal_library_compile_all: compiled 'misc' library in 0.006 sec ggml_metal_library_compile_all: loaded 36 libraries from embedded data in 0.057 sec (max single = 0.056 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": "d327bf1", "build_number": 11869, "cpu_info": "Accelerate, Apple M2 Pro", "gpu_info": "Apple M2 Pro", "backends": "MTL,BLAS", "model_filename": "/Users/minipro/quantdiff/models/campaign/3376b755f9ab/Sage-A1.5.i1-IQ3_M.gguf", "model_type": "qwen35moe 35B.A3B IQ3_S mix - 3.66 bpw", "model_size": 15795632640, "model_n_params": 35505251456, "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": "2026-10-25T20:05:24Z", "avg_ns": 875108766, "stddev_ns": 3945836, "avg_ts": 585.079659, "stddev_ts": 2.641778, "samples_ns": [ 873403666, 869417958, 875594875, 877365417, 879761917 ], "samples_ts": [ 586.212, 588.9, 584.745, 583.565, 581.976 ] }, { "build_commit": "d327bf1", "build_number": 11869, "cpu_info": "Accelerate, Apple M2 Pro", "gpu_info": "Apple M2 Pro", "backends": "MTL,BLAS", "model_filename": "/Users/minipro/quantdiff/models/campaign/3376b755f9ab/Sage-A1.5.i1-IQ3_M.gguf", "model_type": "qwen35moe 35B.A3B IQ3_S mix - 3.66 bpw", "model_size": 15795632640, "model_n_params": 35505251456, "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": "2026-10-25T20:05:29Z", "avg_ns": 2549335700, "stddev_ns": 936278, "avg_ts": 50.209164, "stddev_ts": 0.018438, "samples_ns": [ 2550704334, 2548180333, 2548917083, 2549664334, 2549212416 ], "samples_ts": [ 50.1822, 50.2319, 50.2174, 50.2027, 50.2116 ] } ] # exit 0 after 19s