yaovi78/gemma-3-27b-it-abliterated-gguf overview

{
  "metadata": {},
  "card_data": {
    "license": "gemma",
    "library_name": "transformers",
    "pipeline_tag": "image-text-to-text",
    "base_model": "google/gemma-3-27b-it",
    "tags": [
      "autoquant",
      "gguf"
    ],
    "frontmatter": {
      "license": "gemma",
      "library_name": "transformers",
      "pipeline_tag": "image-text-to-text",
      "base_model": "google/gemma-3-27b-it",
      "tags": [
        "autoquant",
        "gguf"
      ]
    },
    "hero_image_url": "https://cdn-uploads.huggingface.co/production/uploads/61b8e2ba285851687028d395/WjFfc8hhj20r5XK07Yny9.png",
    "summary": "!image/png Gemma 3 4B Abliterated • Gemma 3 12B Abliterated This is an uncensored version of google/gemma-3-27b-it created with a new abliteration technique. See this article to know more about abliteration. I was playing with model weights and noticed that Gemma 3 was much more resilient to abliteration than other models like Qwen 2.5. I experimented with a few recipes to remove refusals while preserving most of the model capabilities. Note that this is fairly experimental, so it might not turn out as well as expected. I recommend using these generation parameters: temperature=1.0, top_k=64, top_p=0.95.",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nlicense: gemma\nlibrary_name: transformers\npipeline_tag: image-text-to-text\nbase_model: google/gemma-3-27b-it\ntags:\n- autoquant\n- gguf\n---\n\n# 💎 Gemma 3 27B IT Abliterated\n\n![image/png](https://cdn-uploads.huggingface.co/production/uploads/61b8e2ba285851687028d395/WjFfc8hhj20r5XK07Yny9.png)\n<center><a href=\"https://huggingface.co/mlabonne/gemma-3-4b-it-abliterated\">Gemma 3 4B Abliterated</a> • <a href=\"https://huggingface.co/mlabonne/gemma-3-12b-it-abliterated\">Gemma 3 12B Abliterated</a></center>\n\nThis is an uncensored version of [google/gemma-3-27b-it](https://huggingface.co/google/gemma-3-27b-it) created with a new abliteration technique.\nSee [this article](https://huggingface.co/blog/mlabonne/abliteration) to know more about abliteration.\n\nI was playing with model weights and noticed that Gemma 3 was much more resilient to abliteration than other models like Qwen 2.5. \nI experimented with a few recipes to remove refusals while preserving most of the model capabilities.\n\nNote that this is fairly experimental, so it might not turn out as well as expected.\n\nI recommend using these generation parameters: `temperature=1.0`, `top_k=64`, `top_p=0.95`.\n\n## ✂️ Layerwise abliteration\n\n![image/png](https://cdn-uploads.huggingface.co/production/uploads/61b8e2ba285851687028d395/HnBRigUfoQaCnpz96jnun.png)\n\nIn the original technique, a refusal direction is computed by comparing the residual streams between target (harmful) and baseline (harmless) samples.\n\nHere, the model was abliterated by computing a refusal direction based on hidden states (inspired by [Sumandora's repo](https://github.com/Sumandora/remove-refusals-with-transformers/)) for each layer, independently.\nThis is combined with a refusal weight of 1.5 to upscale the importance of this refusal direction in each layer.\n\nThis created a very high acceptance rate (>90%) and still produced coherent outputs.\n\n---\n\nThanks to @bartowski for the mmproj file!\n",
    "related_quantizations": []
  },
  "tags": [
    "transformers",
    "gguf",
    "autoquant",
    "image-text-to-text",
    "base_model:google/gemma-3-27b-it",
    "base_model:quantized:google/gemma-3-27b-it",
    "license:gemma",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 444,
  "gated": false,
  "private": false,
  "last_modified": "2026-03-20T11:38:16.000Z",
  "created_at": "2026-03-20T11:38:15.000Z",
  "pipeline_tag": "image-text-to-text",
  "library_name": "transformers"
}

{
  "_id": "69bd31a77a24334f6f6325fe",
  "id": "Yaovi78/gemma-3-27b-it-abliterated-GGUF",
  "modelId": "Yaovi78/gemma-3-27b-it-abliterated-GGUF",
  "sha": "c26d7bf30141d294fe31f09e8a3ed18ca5470598",
  "createdAt": "2026-03-20T11:38:15.000Z",
  "lastModified": "2026-03-20T11:38:16.000Z",
  "author": "Yaovi78",
  "downloads": 444,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "image-text-to-text",
  "library_name": "transformers",
  "siblings_count": 9
}