$ 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-Q2_K_XL.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json
# started 2026-10-21T08:49:12Z
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-Q2_K_XL.gguf",
    "model_type": "qwen35 4B Q2_K - Medium",
    "model_size": 1929857024,
    "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": "2026-10-21T08:49:13Z",
    "avg_ns": 76672420,
    "stddev_ns": 2978976,
    "avg_ts": 6685.486322,
    "stddev_ts": 248.622599,
    "samples_ns": [ 81850000, 74739600, 75539100, 74768000, 76465400 ],
    "samples_ts": [ 6255.35, 6850.45, 6777.95, 6847.85, 6695.84 ]
  },
  {
    "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-Q2_K_XL.gguf",
    "model_type": "qwen35 4B Q2_K - Medium",
    "model_size": 1929857024,
    "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": "2026-10-21T08:49:13Z",
    "avg_ns": 686259060,
    "stddev_ns": 2279249,
    "avg_ts": 186.520131,
    "stddev_ts": 0.619579,
    "samples_ns": [ 685578900, 688097900, 685403400, 688949900, 683265200 ],
    "samples_ts": [ 186.704, 186.02, 186.751, 185.79, 187.336 ]
  }
]

# exit 0 after 5s