$ C:\quantdiff\llama.cpp\bin\llama-bench.exe -m C:\quantdiff\hf-cache\models--mradermacher--Qwen3.5-4B-i1-GGUF\snapshots\8501a8b9bf0a85a86e4abce7b17870bcfad5d128\Qwen3.5-4B.i1-Q4_K_M.gguf -ngl 999 -fa 1 -p 512 -n 128 -r 5 -t 8 -o json
# started 2026-10-21T08:22:54Z
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--Qwen3.5-4B-i1-GGUF\\snapshots\\8501a8b9bf0a85a86e4abce7b17870bcfad5d128\\Qwen3.5-4B.i1-Q4_K_M.gguf",
    "model_type": "qwen35 4B Q4_K - Medium",
    "model_size": 2697836544,
    "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": "2026-10-21T08:22:55Z",
    "avg_ns": 65875100,
    "stddev_ns": 3106942,
    "avg_ts": 7785.331960,
    "stddev_ts": 345.822984,
    "samples_ns": [ 71408700, 64207700, 64261000, 64936300, 64561800 ],
    "samples_ts": [ 7169.99, 7974.12, 7967.51, 7884.65, 7930.39 ]
  },
  {
    "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--Qwen3.5-4B-i1-GGUF\\snapshots\\8501a8b9bf0a85a86e4abce7b17870bcfad5d128\\Qwen3.5-4B.i1-Q4_K_M.gguf",
    "model_type": "qwen35 4B Q4_K - Medium",
    "model_size": 2697836544,
    "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": "2026-10-21T08:22:56Z",
    "avg_ns": 810495360,
    "stddev_ns": 2098771,
    "avg_ts": 157.928955,
    "stddev_ts": 0.407969,
    "samples_ns": [ 813977900, 810895600, 808806600, 809588500, 809208200 ],
    "samples_ts": [ 157.252, 157.85, 158.258, 158.105, 158.179 ]
  }
]

# exit 0 after 6s