aisho67/qwen3.5-4b-opus-4.6-reasoning-distilled-v1-gguf overview

{
  "metadata": {},
  "card_data": {
    "datasets": [
      "crownelius/Opus-4.6-Reasoning-3300x"
    ],
    "base_model": [
      "Qwen/Qwen3.5-4B"
    ],
    "license": "apache-2.0",
    "tags": [
      "qwen",
      "qwen3.5",
      "reasoning"
    ],
    "frontmatter": {
      "datasets": [
        "crownelius/Opus-4.6-Reasoning-3300x"
      ],
      "base_model": [
        "Qwen/Qwen3.5-4B"
      ],
      "license": "apache-2.0",
      "tags": [
        "qwen",
        "qwen3.5",
        "reasoning"
      ]
    },
    "hero_image_url": "",
    "summary": "",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\ndatasets:\n- crownelius/Opus-4.6-Reasoning-3300x\nbase_model:\n- Qwen/Qwen3.5-4B\nlicense: apache-2.0\ntags:\n- qwen\n- qwen3.5\n- reasoning\n---\n\n## Qwen3.5-4b-Opus-4.6-Reasoning-Distilled-v1\nFirst attempt at improving reasoning abilities for Qwen3.5 4b model based on the open [crownelius/Opus-4.6-Reasoning-3300x](https://huggingface.co/datasets/crownelius/Opus-4.6-Reasoning-3300x0) dataset. \n\n## Evals\nI didn't have the patience to run too many evals, but I definitely noticed vibes wise it was a lot more \"opus\"-like (in a good way) in it's reasoning and responses, base Qwen3.5 4b kinda rambles on. Seemed smarter... The one eval I did was was the latest [LiveBench Reasoning benchmarks](https://github.com/LiveBench/LiveBench?tab=readme-ov-file), and here are the results:\n\n| Model | Spatial | Zebra Puzzle | Reasoning Avg |\n|------|--------:|-------------:|--------------:|\n| qwen3.5-4b (Base) | 4.0 | 18.75 | 11.4 |\n| Qwen3.5-4b-Opus-4.6-Reasoning-Distilled-v1 | **24.0** | **19.0** | **21.5** |\n| **Δ Improvement** | +20.0 | +0.25 | +10.1 |\n| **% Improvement** | **+500%** | **+1.3%** | **+88.6%** |\n\n## Notes\nFor v2 of this model, I need to fix the thinking template. It seems like the model ALWAYS does reasoning due to the way I templated the dataset, so I'm doing another training run with explicit think or no think rows. (hopefully that works?).\n\nAlso, I don't know much about training models or ML, I'm a Software Engineer who uses a lot of AI. I just started, and this was pretty much the first real model I've ever trained, so please be nice!\n\n## Training\nI trained this on a single RTX 4060ti with 16GB VRAM, took around 2 or 3 hours.\n\n## Acknowledgements\n- [crownelius](https://huggingface.co/crownelius) for the cleaned dataset\n- [unsloth](https://unsloth.ai/docs/models/qwen3.5#qwen3.5-small-0.8b-2b-4b-9b) for the training arch\n- pewdiepie for inspiring me to try training models (seriously, lol)",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "qwen",
    "qwen3.5",
    "reasoning",
    "dataset:crownelius/Opus-4.6-Reasoning-3300x",
    "base_model:Qwen/Qwen3.5-4B",
    "base_model:quantized:Qwen/Qwen3.5-4B",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 4,
  "downloads": 173,
  "gated": false,
  "private": false,
  "last_modified": "2026-03-05T08:08:05.000Z",
  "created_at": "2026-03-05T07:11:11.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "69a92c8f91260cc543227fd2",
  "id": "Aisho67/Qwen3.5-4b-Opus-4.6-Reasoning-Distilled-v1-GGUF",
  "modelId": "Aisho67/Qwen3.5-4b-Opus-4.6-Reasoning-Distilled-v1-GGUF",
  "sha": "380508e9ac37b8f039bb95bd7641b0b0b4979d2b",
  "createdAt": "2026-03-05T07:11:11.000Z",
  "lastModified": "2026-03-05T08:08:05.000Z",
  "author": "Aisho67",
  "downloads": 173,
  "likes": 4,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 3
}