$ C:\quantdiff\llama.cpp\bin\llama-bench.exe -m C:\quantdiff\hf-cache\models--bartowski--Qwen_Qwen3.5-4B-GGUF\snapshots\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\Qwen_Qwen3.5-4B-IQ3_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json
# started 2026-10-21T08:45:13Z
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 12199 MiB):
  Device 0: NVIDIA GeForce RTX 5070, compute capability 12.0, VMM: yes, VRAM: 12199 MiB
load_backend: loaded CUDA backend from C:\quantdiff\llama.cpp\bin\ggml-cuda.dll
load_backend: loaded RPC backend from C:\quantdiff\llama.cpp\bin\ggml-rpc.dll
load_backend: loaded CPU backend from C:\quantdiff\llama.cpp\bin\ggml-cpu-haswell.dll
[
  {
    "build_commit": "39da130fc",
    "build_number": 11770,
    "cpu_info": "AMD Ryzen 7 5800X 8-Core Processor             ",
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "backends": "CUDA",
    "model_filename": "C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-IQ3_M.gguf",
    "model_type": "qwen35 4B IQ3_S mix - 3.66 bpw",
    "model_size": 2472364032,
    "model_n_params": 4326350848,
    "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": "2026-10-21T08:45:14Z",
    "avg_ns": 62589540,
    "stddev_ns": 3292778,
    "avg_ts": 8197.240589,
    "stddev_ts": 402.978232,
    "samples_ns": [ 68473500, 60899200, 61092100, 61326000, 61156900 ],
    "samples_ts": [ 7477.35, 8407.34, 8380.79, 8348.82, 8371.91 ]
  },
  {
    "build_commit": "39da130fc",
    "build_number": 11770,
    "cpu_info": "AMD Ryzen 7 5800X 8-Core Processor             ",
    "gpu_info": "NVIDIA GeForce RTX 5070",
    "backends": "CUDA",
    "model_filename": "C:\\quantdiff\\hf-cache\\models--bartowski--Qwen_Qwen3.5-4B-GGUF\\snapshots\\4168f45a16a1290d65a4ec0fa312ae917a4c15d6\\Qwen_Qwen3.5-4B-IQ3_M.gguf",
    "model_type": "qwen35 4B IQ3_S mix - 3.66 bpw",
    "model_size": 2472364032,
    "model_n_params": 4326350848,
    "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": "2026-10-21T08:45:15Z",
    "avg_ns": 793096220,
    "stddev_ns": 1477814,
    "avg_ts": 161.393223,
    "stddev_ts": 0.300313,
    "samples_ns": [ 795492700, 792893900, 793202200, 792330300, 791562000 ],
    "samples_ts": [ 160.907, 161.434, 161.371, 161.549, 161.706 ]
  }
]

# exit 0 after 6s