amkkk/qwen3.5-4b-abliterated-aggressive-gguf overview

{
  "metadata": {},
  "card_data": {
    "base_model": [
      "Qwen/Qwen3.5-4B"
    ],
    "pipeline_tag": "image-text-to-text",
    "library_name": "gguf",
    "tags": [
      "gguf",
      "qwen3.5",
      "refusal-direction",
      "model-editing",
      "baked-checkpoint"
    ],
    "license": "apache-2.0",
    "frontmatter": {
      "base_model": [
        "Qwen/Qwen3.5-4B"
      ],
      "pipeline_tag": "image-text-to-text",
      "library_name": "gguf",
      "tags": [
        "gguf",
        "qwen3.5",
        "refusal-direction",
        "model-editing",
        "baked-checkpoint"
      ],
      "license": "apache-2.0"
    },
    "hero_image_url": "",
    "summary": "GGUF release of the baked dense-search candidate based on Qwen/Qwen3.5-4B. This folder contains:",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nbase_model:\n- Qwen/Qwen3.5-4B\npipeline_tag: image-text-to-text\nlibrary_name: gguf\ntags:\n- gguf\n- qwen3.5\n- refusal-direction\n- model-editing\n- baked-checkpoint\nlicense: apache-2.0\n---\n\n# qwen35-4b-oblit-dense-gguf\n\nGGUF release of the baked dense-search candidate based on `Qwen/Qwen3.5-4B`.\n\nThis folder contains:\n\n- a full-fidelity text GGUF\n- a practical spread of text GGUF quants\n- the matching multimodal projector for image support\n\n## Files\n\n- `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.f16.gguf`\n- `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q8_0.gguf`\n- `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q6_K.gguf`\n- `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q5_K_M.gguf`\n- `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q5_K_S.gguf`\n- `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q4_K_M.gguf`\n- `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q4_K_S.gguf`\n- `mmproj-Qwen3.5-4B-F16.gguf`\n\nRecommended default:\n\n- text-only: `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q4_K_M.gguf`\n- multimodal: `qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q4_K_M.gguf` with `mmproj-Qwen3.5-4B-F16.gguf`\n\nQuant guide:\n\n- `F16`: highest fidelity, largest file\n- `Q8_0`: near-lossless quant\n- `Q6_K`: high-quality smaller quant\n- `Q5_K_M`: balanced quality/size\n- `Q5_K_S`: slightly smaller than `Q5_K_M`\n- `Q4_K_M`: practical default\n- `Q4_K_S`: smallest file in this release\n\n## Source candidate\n\n- Baked checkpoint: `../qwen35-4b-oblit-dense-baked`\n- Base model: `Qwen/Qwen3.5-4B`\n- Dense-search edit: layer `20`, strength `1.50`, targets `attention, embeddings, linear_attention, mlp`\n\n## Evaluation snapshot\n\n- Harmful refusals: `128/128 -> 11/128`\n- Harmless refusals: `9/131 -> 5/131`\n- Harmful refusals (Minos): `128/128 -> 28/128`\n- Harmless refusals (Minos): `3/131 -> 4/131`\n- Harmful KL mean: `0.0005979276029393077`\n- Harmless KL mean: `4.5933134970255196e-05`\n\n## Text-Only Usage\n\n```bash\nllama-cli \\\n  -m qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q4_K_M.gguf \\\n  -p \"Explain overfitting in plain English.\"\n```\n\n## Image Usage\n\n```bash\nllama-mtmd-cli \\\n  -m qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q4_K_M.gguf \\\n  --mmproj mmproj-Qwen3.5-4B-F16.gguf \\\n  --image example.jpg \\\n  -p \"Describe this image.\"\n```\n\n## Server Usage\n\n```bash\nllama-server \\\n  -m qwen3.5-4b-obliterate-dense-search-step1-no_decoder_input.Q4_K_M.gguf \\\n  --mmproj mmproj-Qwen3.5-4B-F16.gguf\n```\n\n## Notes\n\n- The `mmproj-Qwen3.5-4B-F16.gguf` file is the matching projector for multimodal use.\n- If you only want text inference, the projector is not required.\n- The `Q4_K_M` file is the practical default. `Q6_K` and `Q8_0` are there if you want to trade space for fidelity.\n- The `F16` file is the highest-fidelity GGUF export in this folder.\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "qwen3.5",
    "refusal-direction",
    "model-editing",
    "baked-checkpoint",
    "image-text-to-text",
    "base_model:Qwen/Qwen3.5-4B",
    "base_model:quantized:Qwen/Qwen3.5-4B",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 1,
  "downloads": 393,
  "gated": false,
  "private": false,
  "last_modified": "2026-03-15T15:09:19.000Z",
  "created_at": "2026-03-13T11:36:22.000Z",
  "pipeline_tag": "image-text-to-text",
  "library_name": "gguf"
}

{
  "_id": "69b3f6b6a76948828f28eff9",
  "id": "amkkk/Qwen3.5-4B-abliterated-aggressive-GGUF",
  "modelId": "amkkk/Qwen3.5-4B-abliterated-aggressive-GGUF",
  "sha": "eb26fa0508e1b7aed235b2afb22d76e17f63f374",
  "createdAt": "2026-03-13T11:36:22.000Z",
  "lastModified": "2026-03-15T15:09:19.000Z",
  "author": "amkkk",
  "downloads": 393,
  "likes": 1,
  "gated": false,
  "private": false,
  "pipeline_tag": "image-text-to-text",
  "library_name": "gguf",
  "siblings_count": 10
}