{
  "exe": "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata/engine/strata",
  "args": [
    "--pack",
    "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata-data/packs/iq3_s",
    "--native",
    "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/llama-cpp-library/models/qwenflashroc-iq3s/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
    "--ple-gguf",
    "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/llama-cpp-library/models/qwenflashroc-iq3s/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00002-of-00002.gguf",
    "--expert-profile",
    "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata/data/expert-profile.bin",
    "--expert-cache",
    "auto",
    "--prefill",
    "1024",
    "--spec",
    "4",
    "--spec-min-p",
    "0.5",
    "--mtp",
    "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata-data/mtp/rt",
    "--max-context",
    "131072",
    "--kv",
    "q4_0",
    "--mmap-experts",
    "--vision",
    "--vram-reserve-mib",
    "700",
    "--kv-resident",
    "32768"
  ],
  "cwd": "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata",
  "tokenizer": "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata-data/packs/iq3_s/tokenizer",
  "model_name": "qwen38-flash-next-iq3s",
  "log": "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata/strata-iq3_s-hermes-dual-gpu.log",
  "lib_dirs": [
    "/usr/local/cuda-13.3/bin",
    "/usr/local/cuda-13.3/lib64"
  ],
  "port": 9931,
  "gpu": [
    0,
    1
  ],
  "gpus_asked": true,
  "vision": {
    "exe": "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/Strata/engine/strata-vision",
    "mmproj": "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/llama-cpp-library/models/qwenflashroc-iq3s/IQ3_S/mmproj-Qwen3.8-Flash-Next-BF16.gguf",
    "model": "/media/daniel/73a99808-dc25-4c9e-be53-2d22b67a035b/llama-cpp-library/models/qwenflashroc-iq3s/IQ3_S/Qwen3.8-Flash-Next-GSQ-RCO-IQ3_S-00001-of-00002.gguf",
    "gpu": false,
    "max_tokens": 300,
    "threads": 12
  },
  "layer_split": "16",
  "host": "0.0.0.0",
  "fit_max_tokens": true,
  "sampling": {
    "temperature": 1.0,
    "top_p": 0.95,
    "top_k": 20,
    "min_p": 0.0,
    "presence_penalty": 0.0,
    "repetition_penalty": 1.0
  }
}