mannix-ita/qwen3.5-27b-omnimerge-gguf overview

{
  "metadata": {},
  "card_data": {
    "base_model": "ManniX-ITA/Qwen3.5-27B-Omnimerge",
    "tags": [
      "gguf",
      "imatrix",
      "quantized",
      "merge",
      "task-arithmetic",
      "qwen3.5",
      "reasoning"
    ],
    "license": "apache-2.0",
    "frontmatter": {
      "base_model": "ManniX-ITA/Qwen3.5-27B-Omnimerge",
      "tags": [
        "gguf",
        "imatrix",
        "quantized",
        "merge",
        "task-arithmetic",
        "qwen3.5",
        "reasoning"
      ],
      "license": "apache-2.0"
    },
    "hero_image_url": "",
    "summary": "GGUF quantizations of ManniX-ITA/Qwen3.5-27B-Omnimerge — a 3-way **Task Arithmetic** weight-space merge of three Qwen3.5-27B reasoning-distilled fine-tunes. **This merge outperforms its best source model (Claude-4.6-Opus-Reasoning-Distilled) across all tested benchmarks: +8 pp on GPQA Diamond reasoning, +3.7 pp on HumanEval, and comparable MBPP.** All quants made with imatrix using calibration data v5.",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nbase_model: ManniX-ITA/Qwen3.5-27B-Omnimerge\ntags:\n  - gguf\n  - imatrix\n  - quantized\n  - merge\n  - task-arithmetic\n  - qwen3.5\n  - reasoning\nlicense: apache-2.0\n---\n\n# Qwen3.5-27B-Omnimerge-GGUF\n\nGGUF quantizations of [ManniX-ITA/Qwen3.5-27B-Omnimerge](https://huggingface.co/ManniX-ITA/Qwen3.5-27B-Omnimerge) — a 3-way **Task Arithmetic** weight-space merge of three Qwen3.5-27B reasoning-distilled fine-tunes.\n\n**This merge outperforms its best source model (Claude-4.6-Opus-Reasoning-Distilled) across all tested benchmarks: +8 pp on GPQA Diamond reasoning, +3.7 pp on HumanEval, and comparable MBPP.**\n\nAll quants made with imatrix using [calibration data v5](https://gist.github.com/bartowski1182/82ae9b520227f57d79ba04add13d0d0d).\n\n## Benchmark Results (Q6_K)\n\n| Benchmark | Omnimerge | Claude-distill (best source) | Delta |\n|---|---|---|---|\n| **GPQA Diamond** (198q, flex) | **61.11%** | 53.03% | **+8.08 pp** |\n| **HumanEval** pass@1 | **79.88%** | 76.22% | **+3.66 pp** |\n| **MBPP** pass@1 | **71.80%** | 71.20% | **+0.60 pp** |\n\n## Available Quantizations\n\n| Quantization | File | Size |\n|---|---|---|\n| Q8_0 | merged_omnimerge-Q8_0.gguf | 26.63 GB |\n| Q6_K_L | merged_omnimerge-Q6_K_L.gguf | 21.14 GB |\n| Q6_K | merged_omnimerge-Q6_K.gguf | 20.57 GB |\n| Q5_K_L | merged_omnimerge-Q5_K_L.gguf | 18.64 GB |\n| Q5_K_M | merged_omnimerge-Q5_K_M.gguf | 17.91 GB |\n| Q5_K_S | merged_omnimerge-Q5_K_S.gguf | 17.40 GB |\n| Q4_K_L | merged_omnimerge-Q4_K_L.gguf | 16.29 GB |\n| Q4_1 | merged_omnimerge-Q4_1.gguf | 15.91 GB |\n| Q4_K_M | merged_omnimerge-Q4_K_M.gguf | 15.41 GB |\n| IQ4_NL | merged_omnimerge-IQ4_NL.gguf | 14.72 GB |\n| Q4_K_S | merged_omnimerge-Q4_K_S.gguf | 14.52 GB |\n| Q4_0 | merged_omnimerge-Q4_0.gguf | 14.41 GB |\n| IQ4_XS | merged_omnimerge-IQ4_XS.gguf | 14.05 GB |\n| Q3_K_XL | merged_omnimerge-Q3_K_XL.gguf | 13.42 GB |\n| Q3_K_L | merged_omnimerge-Q3_K_L.gguf | 13.36 GB |\n| Q3_K_M | merged_omnimerge-Q3_K_M.gguf | 12.39 GB |\n| IQ3_M | merged_omnimerge-IQ3_M.gguf | 11.72 GB |\n| Q3_K_S | merged_omnimerge-Q3_K_S.gguf | 11.24 GB |\n| IQ3_XS | merged_omnimerge-IQ3_XS.gguf | 11.15 GB |\n| IQ3_XXS | merged_omnimerge-IQ3_XXS.gguf | 10.42 GB |\n\n### Skipped Quantizations (failed sanity check)\n\nThe following 2-bit quantizations were attempted but **failed the sanity check** (3 capital city questions answered incorrectly or incoherently). At 2-bit precision on a 27B model, too much information is lost for reliable output. These quants are intentionally not published:\n\n- Q2_K_L, Q2_K, IQ2_M, IQ2_S, IQ2_XS, IQ2_XXS\n\n## Recommended Usage\n\n```bash\nllama-server -m Qwen3.5-27B-Omnimerge-Q6_K.gguf -c 32768 -ngl 99 \\\n    --jinja --reasoning-format deepseek --reasoning-budget 16384 \\\n    --temp 0.6 --top-p 0.95 --top-k 20 --dry-multiplier 0.5\n```\n\nFor code tasks, use without `--jinja --reasoning-format deepseek` (plain completions mode).\n\n## Source Models\n\n| Source | Weight | Focus |\n|---|---|---|\n| [Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled](https://huggingface.co/Jackrong/Qwen3.5-27B-Claude-4.6-Opus-Reasoning-Distilled) | 0.40 | Claude 4.6 Opus reasoning distillation |\n| [ValiantLabs/Qwen3.5-27B-Esper3.1](https://huggingface.co/ValiantLabs/Qwen3.5-27B-Esper3.1) | 0.35 | Code / DevOps specialist |\n| [DavidAU/Qwen3.5-27B-Gemini3-Pro-High-Reasoning-Compact-Thinking](https://huggingface.co/DavidAU/Qwen3.5-27B-Gemini3-Pro-High-Reasoning-Compact-Thinking) | 0.25 | Gemini 3 Pro reasoning, compact thinking |\n\n**Base**: [Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B)\n\nSee the [model card](https://huggingface.co/ManniX-ITA/Qwen3.5-27B-Omnimerge) for full methodology, evaluation details, and the custom merger script.\n\n## License\n\nApache-2.0\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "imatrix",
    "quantized",
    "merge",
    "task-arithmetic",
    "qwen3.5",
    "reasoning",
    "base_model:ManniX-ITA/Qwen3.5-27B-Omnimerge",
    "base_model:quantized:ManniX-ITA/Qwen3.5-27B-Omnimerge",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 3769,
  "gated": false,
  "private": false,
  "last_modified": "2026-04-13T06:02:08.000Z",
  "created_at": "2026-04-12T12:38:52.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "69db925ccf8e40febe4c5102",
  "id": "ManniX-ITA/Qwen3.5-27B-Omnimerge-GGUF",
  "modelId": "ManniX-ITA/Qwen3.5-27B-Omnimerge-GGUF",
  "sha": "655b50fa79811cd79ab2e47af499a3f4e7b2fd17",
  "createdAt": "2026-04-12T12:38:52.000Z",
  "lastModified": "2026-04-13T06:02:08.000Z",
  "author": "ManniX-ITA",
  "downloads": 3769,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 23
}