{
  "_comment": "034 part 2 (2026-08-27 ~01:00, Jose: 'can we go up a quant and fit it'): UD-Q3_K_XL (90 GB) and UD-IQ4_XS (93.7 GB), same protocol as part 1, same output dir so the page shows all columns. Expected peaks ~98 and ~101 GB on 123 GB usable; UD-Q4_K_XL (111 GB) is the cliff and is not attempted.",
  "port": 8899,
  "output_dir": "runs/flash-next-034",
  "max_tokens": 120000,
  "seed": 42,
  "request_timeout_minutes": 240,
  "tests": [
    "01-blockfall",
    "03-eruption",
    "04-the-ledger",
    "05-blind-artist"
  ],
  "contestants": [
    {
      "id": "flashnext-q3kxl-think-medium",
      "label": "Qwen 3.8-Flash-Next UD-Q3_K_XL (90.0 GB) THINKING medium (card samplers)",
      "type": "llamacpp",
      "repo": "unsloth/Qwen3.8-Flash-Next-GGUF",
      "variant": "UD-Q3_K_XL",
      "vision": false,
      "launch_chat_template_kwargs": {
        "enable_thinking": true,
        "reasoning_effort": "medium"
      },
      "sampling": {
        "temperature": 1.0,
        "top_p": 0.95,
        "top_k": 20,
        "min_p": 0,
        "presence_penalty": 0,
        "repeat_penalty": 1.0
      },
      "max_seq_length": 131072,
      "binary": "~/private/jose-yt/tools/llama.cpp-qwen4exp/build/bin/llama-server",
      "extra_launch_args": [
        "--flash-attn",
        "on"
      ]
    },
    {
      "id": "flashnext-iq4xs-think-medium",
      "label": "Qwen 3.8-Flash-Next UD-IQ4_XS (93.7 GB) THINKING medium (card samplers)",
      "type": "llamacpp",
      "repo": "unsloth/Qwen3.8-Flash-Next-GGUF",
      "variant": "UD-IQ4_XS",
      "vision": false,
      "launch_chat_template_kwargs": {
        "enable_thinking": true,
        "reasoning_effort": "medium"
      },
      "sampling": {
        "temperature": 1.0,
        "top_p": 0.95,
        "top_k": 20,
        "min_p": 0,
        "presence_penalty": 0,
        "repeat_penalty": 1.0
      },
      "max_seq_length": 131072,
      "binary": "~/private/jose-yt/tools/llama.cpp-qwen4exp/build/bin/llama-server",
      "extra_launch_args": [
        "--flash-attn",
        "on"
      ]
    }
  ]
}