$ C:\quantdiff\llama.cpp\bin\llama-bench.exe -m C:\quantdiff\hf-cache\models--mradermacher--Qwen3.5-9B-i1-GGUF\snapshots\c5044c600a7e15246d54d13e4532382bc44ef41b\Qwen3.5-9B.i1-IQ4_XS.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json
# started 2026-10-21T06:09:06Z
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--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-IQ4_XS.gguf",
    "model_type": "qwen35 9B IQ4_XS - 4.25 bpw",
    "model_size": 5059643392,
    "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": "2026-10-21T06:09:08Z",
    "avg_ns": 96186740,
    "stddev_ns": 3096029,
    "avg_ts": 5327.212449,
    "stddev_ts": 164.404390,
    "samples_ns": [ 101720300, 94850700, 94921100, 94585700, 94855900 ],
    "samples_ts": [ 5033.41, 5397.96, 5393.95, 5413.08, 5397.66 ]
  },
  {
    "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--mradermacher--Qwen3.5-9B-i1-GGUF\\snapshots\\c5044c600a7e15246d54d13e4532382bc44ef41b\\Qwen3.5-9B.i1-IQ4_XS.gguf",
    "model_type": "qwen35 9B IQ4_XS - 4.25 bpw",
    "model_size": 5059643392,
    "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": "2026-10-21T06:09:08Z",
    "avg_ns": 1126631760,
    "stddev_ns": 1260029,
    "avg_ts": 113.613101,
    "stddev_ts": 0.126989,
    "samples_ns": [ 1127445200, 1125764500, 1125946000, 1128449500, 1125553600 ],
    "samples_ts": [ 113.531, 113.701, 113.682, 113.43, 113.722 ]
  }
]

# exit 0 after 9s