mudler/qwen3.5-397b-a17b-apex-gguf overview

{
  "metadata": {},
  "card_data": {
    "license": "apache-2.0",
    "base_model": "Qwen/Qwen3.5-397B-A17B",
    "tags": [
      "gguf",
      "quantized",
      "apex",
      "moe",
      "mixture-of-experts",
      "qwen3.5",
      "vlm",
      "vision"
    ],
    "frontmatter": {
      "license": "apache-2.0",
      "base_model": "Qwen/Qwen3.5-397B-A17B",
      "tags": [
        "gguf",
        "quantized",
        "apex",
        "moe",
        "mixture-of-experts",
        "qwen3.5",
        "vlm",
        "vision"
      ]
    },
    "hero_image_url": "",
    "summary": "**APEX (Adaptive Precision for EXpert Models)** quantizations of Qwen3.5-397B-A17B. **Brought to you by the LocalAI team** | APEX Project | Technical Report",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nlicense: apache-2.0\nbase_model: Qwen/Qwen3.5-397B-A17B\ntags:\n  - gguf\n  - quantized\n  - apex\n  - moe\n  - mixture-of-experts\n  - qwen3.5\n  - vlm\n  - vision\n---\n\n# Qwen3.5-397B-A17B APEX GGUF\n\n**APEX (Adaptive Precision for EXpert Models)** quantizations of [Qwen3.5-397B-A17B](https://huggingface.co/Qwen/Qwen3.5-397B-A17B).\n\n**Brought to you by the [LocalAI](https://github.com/mudler/LocalAI) team** | [APEX Project](https://github.com/mudler/apex-quant) | [Technical Report](https://github.com/mudler/apex-quant/blob/main/paper/APEX_Technical_Report.pdf)\n\n## Benchmark Results\n\nBenchmarks coming soon. For reference APEX benchmarks on the Qwen3.5-35B-A3B architecture, see [mudler/Qwen3.5-35B-A3B-APEX-GGUF](https://huggingface.co/mudler/Qwen3.5-35B-A3B-APEX-GGUF).\n\n## What is APEX?\n\nAPEX is a quantization strategy for Mixture-of-Experts (MoE) models. It classifies tensors by role (routed expert, shared expert, attention) and applies a layer-wise precision gradient -- edge layers get higher precision, middle layers get more aggressive compression. I-variants use diverse imatrix calibration (chat, code, reasoning, tool-calling, agentic traces, Wikipedia).\n\nSee the [APEX project](https://github.com/mudler/apex-quant) for full details, technical report, and scripts.\n\n## Architecture\n\n- **Model**: Qwen3.5-397B-A17B (qwen3_5_moe)\n- **Layers**: 60 (hybrid: linear attention + full attention every 4th layer)\n- **Experts**: 512 routed (10 active per token)\n- **Total Parameters**: ~397B\n- **Active Parameters**: ~17B per token\n- **Vision**: Built-in vision encoder (mmproj included)\n- **Context**: 262K tokens\n- **APEX Config**: 5+5 symmetric edge gradient across 60 layers\n- **Calibration**: v1.3 diverse dataset (chat, code, reasoning, multilingual, tool-calling, Wikipedia)\n\n## Run with LocalAI\n\n```bash\nlocal-ai run mudler/Qwen3.5-397B-A17B-APEX-GGUF@Qwen3.5-397B-A17B-APEX-I-Balanced.gguf\n```\n\n## Credits\n\nAPEX is brought to you by the [LocalAI](https://github.com/mudler/LocalAI) team. Developed through human-driven, AI-assisted research. Built on [llama.cpp](https://github.com/ggerganov/llama.cpp).\n",
    "related_quantizations": []
  },
  "tags": [
    "gguf",
    "quantized",
    "apex",
    "moe",
    "mixture-of-experts",
    "qwen3.5",
    "vlm",
    "vision",
    "base_model:Qwen/Qwen3.5-397B-A17B",
    "base_model:quantized:Qwen/Qwen3.5-397B-A17B",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 3,
  "downloads": 1758,
  "gated": false,
  "private": false,
  "last_modified": "2026-04-06T07:31:50.000Z",
  "created_at": "2026-04-06T04:01:36.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "69d3302059ede2545f926d89",
  "id": "mudler/Qwen3.5-397B-A17B-APEX-GGUF",
  "modelId": "mudler/Qwen3.5-397B-A17B-APEX-GGUF",
  "sha": "b6584c5959d603c4e26707ab063774d9779dcdf9",
  "createdAt": "2026-04-06T04:01:36.000Z",
  "lastModified": "2026-04-06T07:31:50.000Z",
  "author": "mudler",
  "downloads": 1758,
  "likes": 3,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 6
}