$ C:\quantdiff\llama.cpp\bin\llama-bench.exe -m C:\quantdiff\hf-cache\models--unsloth--Qwen3.5-9B-GGUF\snapshots\3885219b6810b007914f3a7950a8d1b469d598a5\Qwen3.5-9B-Q8_0.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json
# started 2026-10-21T06:15:27Z
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--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q8_0.gguf",
    "model_type": "qwen35 9B Q8_0",
    "model_size": 9516533760,
    "model_n_params": 8953803264,
    "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": "2026-10-21T06:15:31Z",
    "avg_ns": 102778080,
    "stddev_ns": 3154545,
    "avg_ts": 4985.218313,
    "stddev_ts": 147.078734,
    "samples_ns": [ 108399000, 101016100, 101300300, 101795500, 101379500 ],
    "samples_ts": [ 4723.29, 5068.5, 5054.28, 5029.69, 5050.33 ]
  },
  {
    "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--unsloth--Qwen3.5-9B-GGUF\\snapshots\\3885219b6810b007914f3a7950a8d1b469d598a5\\Qwen3.5-9B-Q8_0.gguf",
    "model_type": "qwen35 9B Q8_0",
    "model_size": 9516533760,
    "model_n_params": 8953803264,
    "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": "2026-10-21T06:15:32Z",
    "avg_ns": 1926731480,
    "stddev_ns": 5667549,
    "avg_ts": 66.434210,
    "stddev_ts": 0.194916,
    "samples_ns": [ 1927127600, 1924047800, 1925119200, 1936128600, 1921234200 ],
    "samples_ts": [ 66.4201, 66.5264, 66.4894, 66.1113, 66.6238 ]
  }
]

# exit 0 after 14s