$ C:\quantdiff\llama.cpp\bin\llama-bench.exe -m C:\quantdiff\hf-cache\models--unsloth--Qwen3.5-4B-GGUF\snapshots\e87f176479d0855a907a41277aca2f8ee7a09523\Qwen3.5-4B-UD-Q4_K_XL.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json
# started 2026-10-21T08:29:05Z
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--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q4_K_XL.gguf",
    "model_type": "qwen35 4B Q4_K - Medium",
    "model_size": 2901141504,
    "model_n_params": 4205751296,
    "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": "2026-10-21T08:29:06Z",
    "avg_ns": 64692580,
    "stddev_ns": 3117684,
    "avg_ts": 7928.215627,
    "stddev_ts": 359.553628,
    "samples_ns": [ 70237200, 62964500, 63141000, 63261500, 63858700 ],
    "samples_ts": [ 7289.58, 8131.57, 8108.84, 8093.39, 8017.7 ]
  },
  {
    "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--Qwen3.5-4B-GGUF\\snapshots\\e87f176479d0855a907a41277aca2f8ee7a09523\\Qwen3.5-4B-UD-Q4_K_XL.gguf",
    "model_type": "qwen35 4B Q4_K - Medium",
    "model_size": 2901141504,
    "model_n_params": 4205751296,
    "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": "2026-10-21T08:29:07Z",
    "avg_ns": 850001480,
    "stddev_ns": 10352343,
    "avg_ts": 150.605741,
    "stddev_ts": 1.823606,
    "samples_ns": [ 864244700, 857848200, 843627000, 841896900, 842390600 ],
    "samples_ts": [ 148.106, 149.211, 151.726, 152.038, 151.949 ]
  }
]

# exit 0 after 7s