{
  "items": [
    {
      "label": {
        "en": "Qwen3-8B Q4_K_M — Model on GPU",
        "pl": "Qwen3-8B Q4_K_M — Model na GPU"
      },
      "value": 3.119277758589535
    },
    {
      "label": {
        "en": "Qwen3-32B Q4_K_M — Model on GPU",
        "pl": "Qwen3-32B Q4_K_M — Model na GPU"
      },
      "value": 3.1198924303190445
    },
    {
      "label": {
        "en": "Qwen2.5-72B-Instruct Q4_K_M — Model on GPU",
        "pl": "Qwen2.5-72B-Instruct Q4_K_M — Model na GPU"
      },
      "value": 3.1211448738400773
    },
    {
      "label": {
        "en": "Qwen2.5-72B-Instruct Q4_K_M — GPU + CPU: lower VRAM use",
        "pl": "Qwen2.5-72B-Instruct Q4_K_M — GPU + CPU: mniejsze zajęcie VRAM"
      },
      "value": 46.961556060353814
    },
    {
      "label": {
        "en": "Qwen2.5-72B-Instruct Q4_K_M — GPU + CPU: further reduced VRAM use",
        "pl": "Qwen2.5-72B-Instruct Q4_K_M — GPU + CPU: jeszcze mniejsze zajęcie VRAM"
      },
      "value": 47.48170472432676
    }
  ],
  "note": {
    "en": "Process share of the server’s 32 logical CPUs, weighted by measured interval duration. 2048 input tokens, 5 rounds, one request at a time; not all-service CPU load.",
    "pl": "Udział procesu względem 32 wątków serwera, ważony czasem pomiaru. Wejście 2048 tokenów, 5 rund, jedno zapytanie naraz; nie obciążenie wszystkich usług."
  },
  "title": {
    "en": "CPU used by the model",
    "pl": "CPU wykorzystywany przez model"
  },
  "unit": "%"
}