empero-ai/qwen3.5-9b-claude-opus-4.6-distill-gguf overview

{
  "metadata": {},
  "card_data": {
    "license": "apache-2.0",
    "language": [
      "en"
    ],
    "base_model": [
      "Qwen/Qwen3.5-9B"
    ],
    "library_name": "transformers",
    "tags": [
      "qwen",
      "claude",
      "opus",
      "reasoning",
      "distill"
    ],
    "datasets": [
      "nohurry/Opus-4.6-Reasoning-3000x-filtered",
      "Jackrong/Qwen3.5-reasoning-700x",
      "TeichAI/claude-4.5-opus-high-reasoning-250x"
    ],
    "frontmatter": {
      "license": "apache-2.0",
      "language": [
        "en"
      ],
      "base_model": [
        "Qwen/Qwen3.5-9B"
      ],
      "library_name": "transformers",
      "tags": [
        "qwen",
        "claude",
        "opus",
        "reasoning",
        "distill"
      ],
      "datasets": [
        "nohurry/Opus-4.6-Reasoning-3000x-filtered",
        "Jackrong/Qwen3.5-reasoning-700x",
        "TeichAI/claude-4.5-opus-high-reasoning-250x"
      ]
    },
    "hero_image_url": "",
    "summary": "# Qwen3.5-9B Claude Opus 4.6 Reasoning Distill — GGUF GGUF quantizations of empero-ai/Qwen3.5-9B-Claude-Opus-4.6-Distill, a reasoning-focused fine-tune of Qwen/Qwen3.5-9B. This model was trained to produce detailed chain-of-thought reasoning inside  tags before giving its final answer, distilled from Claude Opus 4.6 and Qwen3.5 reasoning traces.",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nlicense: apache-2.0\nlanguage:\n- en\nbase_model:\n- Qwen/Qwen3.5-9B\nlibrary_name: transformers\ntags:\n- qwen\n- claude\n- opus\n- reasoning\n- distill\ndatasets:\n- nohurry/Opus-4.6-Reasoning-3000x-filtered\n- Jackrong/Qwen3.5-reasoning-700x\n- TeichAI/claude-4.5-opus-high-reasoning-250x\n---\n# Qwen3.5-9B Claude Opus 4.6 Reasoning Distill — GGUF\n \nGGUF quantizations of [empero-ai/Qwen3.5-9B-Claude-Opus-4.6-Distill](https://huggingface.co/empero-ai/Qwen3.5-9B-Claude-Opus-4.6-Distill), a reasoning-focused fine-tune of [Qwen/Qwen3.5-9B](https://huggingface.co/Qwen/Qwen3.5-9B).\n \nThis model was trained to produce detailed chain-of-thought reasoning inside `<think>` tags before giving its final answer, distilled from Claude Opus 4.6 and Qwen3.5 reasoning traces.\n \n## Quantizations\n \n| File | Quant | Size | Description |\n|------|-------|------|-------------|\n| `qwen3.5-9b-opus4.6-distill-Q2_K.gguf` | Q2_K | ~3.5 GB | Smallest, lowest quality. For very constrained devices. |\n| `qwen3.5-9b-opus4.6-distill-Q3_K_M.gguf` | Q3_K_M | ~4.5 GB | Low quality, usable for testing. |\n| `qwen3.5-9b-opus4.6-distill-Q4_K_M.gguf` | Q4_K_M | ~5.5 GB | **Recommended.** Best balance of quality and size. |\n| `qwen3.5-9b-opus4.6-distill-Q5_K_M.gguf` | Q5_K_M | ~6.5 GB | High quality, moderate size. |\n| `qwen3.5-9b-opus4.6-distill-Q6_K.gguf` | Q6_K | ~7.5 GB | Very high quality, near-lossless. |\n| `qwen3.5-9b-opus4.6-distill-Q8_0.gguf` | Q8_0 | ~9.5 GB | Highest quality quantization. |\n| `qwen3.5-9b-opus4.6-distill-f16.gguf` | F16 | ~18 GB | Full precision, no quantization loss. |\n \nFor most users, **Q4_K_M** or **Q5_K_M** is the sweet spot.\n \n## How to Use\n \n### llama.cpp\n \n```bash\nllama-cli -m qwen3.5-9b-opus4.6-distill-Q5_K_M.gguf -p \"<|im_start|>system\\nYou are a deep reasoning AI. Think carefully inside <think> tags before answering.<|im_end|>\\n<|im_start|>user\\nExplain why the sky is blue.<|im_end|>\\n<|im_start|>assistant\\n\" -n 2048\n```\n \n### Ollama\n \n```bash\nollama run empero-ai/qwen3.5-9b-opus4.6-distill\n```\n \n### LM Studio / GPT4All / Jan\n \nDownload the GGUF file of your choice and load it directly in the application.\n \n## Training Details\n \n### Method\n \n- **Stage 1 — SFT (Supervised Fine-Tuning):** 3 epochs on ~13K examples teaching the model the `<think>` reasoning format using QLoRA (4-bit, rank 64, alpha 128)\n- **Base model:** Qwen/Qwen3.5-9B\n- **Hardware:** RTX 5090 (32GB VRAM)\n- **Attention:** SDPA\n- **Optimizer:** Paged AdamW 8-bit\n- **Learning rate:** 1e-4 with cosine schedule\n- **Effective batch size:** 8 (batch 1 × gradient accumulation 8)\n- **Max sequence length:** 4096\n \n### SFT Results\n \n| Metric | Epoch 1 | Epoch 2 (best) | Epoch 3 |\n|--------|---------|-----------------|---------|\n| Eval Loss | 0.5205 | **0.4809** | 0.4915 |\n| Eval Token Accuracy | 0.8494 | **0.8615** | 0.8617 |\n| Eval Entropy | 0.508 | 0.434 | 0.394 |\n \nBest checkpoint (epoch 2) was selected via `load_best_model_at_end`.\n \n### Datasets\n \n| Dataset | Examples | Type |\n|---------|----------|------|\n| [nohurry/Opus-4.6-Reasoning-3000x-filtered](https://huggingface.co/datasets/nohurry/Opus-4.6-Reasoning-3000x-filtered) | 2,326 | Problem → thinking → solution |\n| [Jackrong/Qwen3.5-reasoning-700x](https://huggingface.co/datasets/Jackrong/Qwen3.5-reasoning-700x) | 633 | ShareGPT with `<think>` tags |\n| [TeichAI/claude-4.5-opus-high-reasoning-250x](https://huggingface.co/datasets/TeichAI/claude-4.5-opus-high-reasoning-250x) | 250 | Messages with `<think>` tags |\n| [Roman1111111/claude-opus-4.6-10000x](https://huggingface.co/datasets/Roman1111111/claude-opus-4.6-10000x) | 9,631 | Messages with reasoning traces |\n| **Total** | **12,840** | |\n \n### Output Format\n \nThe model outputs reasoning in `<think>` tags followed by its final answer:\n \n```\n<think>\nThe user is asking about why the sky appears blue. This involves Rayleigh scattering...\n \nSunlight contains all wavelengths of visible light. When it enters Earth's atmosphere,\nshorter wavelengths (blue/violet) scatter more than longer wavelengths (red/orange)...\n \nWhile violet actually scatters more than blue, our eyes are more sensitive to blue light,\nand some violet is absorbed by the upper atmosphere...\n</think>\n \nThe sky appears blue due to Rayleigh scattering. When sunlight passes through Earth's\natmosphere, the shorter blue wavelengths scatter in all directions more than the longer\nred wavelengths. Although violet light scatters even more, our eyes are more sensitive\nto blue, and some violet is absorbed higher in the atmosphere — so we perceive the sky\nas blue.\n```\n\n## About Empero AI\n \nThis model was developed by [Empero AI](https://empero.org). We build open-source AI tools and models focused on advancing reasoning capabilities in smaller, efficient language models.\n \n## License\n \nThis model inherits the [Apache 2.0](https://www.apache.org/licenses/LICENSE-2.0) license from Qwen3.5-9B.",
    "related_quantizations": []
  },
  "tags": [
    "transformers",
    "gguf",
    "qwen",
    "claude",
    "opus",
    "reasoning",
    "distill",
    "en",
    "dataset:nohurry/Opus-4.6-Reasoning-3000x-filtered",
    "dataset:Jackrong/Qwen3.5-reasoning-700x",
    "dataset:TeichAI/claude-4.5-opus-high-reasoning-250x",
    "base_model:Qwen/Qwen3.5-9B",
    "base_model:quantized:Qwen/Qwen3.5-9B",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 0,
  "downloads": 629,
  "gated": false,
  "private": false,
  "last_modified": "2026-03-15T23:23:35.000Z",
  "created_at": "2026-03-15T22:56:16.000Z",
  "pipeline_tag": "",
  "library_name": "transformers"
}

{
  "_id": "69b739107ea817dbad1988b7",
  "id": "empero-ai/Qwen3.5-9B-Claude-Opus-4.6-Distill-GGUF",
  "modelId": "empero-ai/Qwen3.5-9B-Claude-Opus-4.6-Distill-GGUF",
  "sha": "892a4362ba5896d67d566a316973fefb70f5135f",
  "createdAt": "2026-03-15T22:56:16.000Z",
  "lastModified": "2026-03-15T23:23:35.000Z",
  "author": "empero-ai",
  "downloads": 629,
  "likes": 0,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "transformers",
  "siblings_count": 9
}