mudler/qwen3.5-122b-a10b-apex-gguf overview

{
  "metadata": {},
  "card_data": {
    "license": "apache-2.0",
    "base_model": "Qwen/Qwen3.5-122B-A10B",
    "tags": [
      "gguf",
      "quantized",
      "apex",
      "moe",
      "mixture-of-experts",
      "qwen3.5"
    ],
    "frontmatter": {
      "license": "apache-2.0",
      "base_model": "Qwen/Qwen3.5-122B-A10B",
      "tags": [
        "gguf",
        "quantized",
        "apex",
        "moe",
        "mixture-of-experts",
        "qwen3.5"
      ]
    },
    "hero_image_url": "",
    "summary": "**APEX (Adaptive Precision for EXpert Models)** quantizations of Qwen3.5-122B-A10B. **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/Qwen3.5-122B-A10B\ntags:\n  - gguf\n  - quantized\n  - apex\n  - moe\n  - mixture-of-experts\n  - qwen3.5\n---\n\n# Qwen3.5-122B-A10B APEX GGUF\n\n**APEX (Adaptive Precision for EXpert Models)** quantizations of [Qwen3.5-122B-A10B](https://huggingface.co/Qwen/Qwen3.5-122B-A10B).\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\nAll measurements on 8xRTX PRO 6000 Blackwell (768 GB VRAM). Perplexity on wikitext-2-raw, context 512. Accuracy benchmarks via llama.cpp (400 tasks each).\n\n| Configuration | Size (GB) | Perplexity | KL mean | HellaSwag | Winogrande | MMLU | ARC | tg128 (t/s) |\n|---------------|-----------|-----------|---------|-----------|------------|------|-----|-------------|\n| Q8_0 (Unsloth) | 121 | 4.819 | 0.004 | 85.5% | 77.3% | 44.19 | 57.19 | 85.5 |\n| Q5_K_S (Unsloth) | ~81 | 4.826 | 0.007 | 85.3% | 76.0% | 43.80 | 57.86 | 90.4 |\n| UD-Q4_K_XL (Unsloth) | ~72 | 4.829 | 0.010 | 84.8% | 76.3% | 44.25 | 55.85 | 91.8 |\n| **APEX I-Balanced** | **83.4** | **4.831** | **0.008** | **85.5%** | **77.8%** | 43.86 | **57.86** | **96.7** |\n| **APEX I-Quality** | **72.3** | **4.838** | 0.012 | 85.3% | 77.3% | 43.86 | 56.86 | **99.7** |\n| APEX Quality | 72.3 | 4.848 | 0.013 | **85.5%** | 76.3% | **44.44** | 55.52 | 99.8 |\n| APEX Balanced | 83.4 | 4.840 | 0.008 | 85.0% | 76.3% | 43.93 | 56.86 | 96.7 |\n| **APEX I-Compact** | **55.1** | **4.978** | 0.041 | 84.5% | 77.5% | **44.06** | **57.86** | **106.3** |\n| APEX Compact | 55.1 | 5.046 | 0.049 | 84.5% | **77.8%** | 43.54 | 56.19 | 106.2 |\n| **APEX I-Mini** | **44.9** | **5.306** | 0.102 | 84.0% | 75.3% | 42.83 | 56.52 | **110.0** |\n\n### Highlights\n\n- **APEX I-Balanced matches or beats Q8_0** on HellaSwag (85.5%), Winogrande (77.8% vs 77.3%), and ARC (57.86 vs 57.19) while being **31% smaller** and **13% faster**.\n- **APEX I-Quality (72.3 GB)** beats UD-Q4_K_XL at the same size on HellaSwag (85.3% vs 84.8%), Winogrande (77.3% vs 76.3%), and ARC (56.86 vs 55.85).\n- **APEX I-Compact (55.1 GB)** achieves 84.5% HellaSwag and 57.86 ARC at **55% less size than Q8_0** — fastest standard profile at 106 t/s.\n- **APEX I-Mini (44.9 GB)** is the smallest at **63% less size than Q8_0**, still 84% HellaSwag, fastest at 110 t/s.\n- I-variants consistently improve over standard profiles across PPL, KL, and ARC.\n\n## Available Files\n\n| File | Profile | Size | Best For |\n|------|---------|------|----------|\n| Qwen3.5-122B-A10B-APEX-I-Balanced.gguf | I-Balanced | 83.4 GB | Best overall -- matches Q8_0 quality at 31% less size |\n| Qwen3.5-122B-A10B-APEX-I-Quality.gguf | I-Quality | 72.3 GB | Best quality at ~72 GB tier |\n| Qwen3.5-122B-A10B-APEX-Quality.gguf | Quality | 72.3 GB | Highest MMLU (44.44) |\n| Qwen3.5-122B-A10B-APEX-Balanced.gguf | Balanced | 83.4 GB | General purpose, low KL |\n| Qwen3.5-122B-A10B-APEX-I-Compact.gguf | I-Compact | 55.1 GB | Consumer multi-GPU, best quality/size ratio |\n| Qwen3.5-122B-A10B-APEX-Compact.gguf | Compact | 55.1 GB | Consumer multi-GPU setups |\n| Qwen3.5-122B-A10B-APEX-I-Mini.gguf | I-Mini | 44.9 GB | Smallest viable, fastest inference |\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**: Qwen3.5-122B-A10B (Qwen3.5-MoE)\n- **Layers**: 48\n- **Experts**: 256 routed + 1 shared (8 active per token)\n- **Total Parameters**: 122B\n- **Active Parameters**: ~10B per token\n- **APEX Config**: 5+5 symmetric edge gradient across 48 layers\n\n## Run with LocalAI\n\n```bash\nlocal-ai run mudler/Qwen3.5-122B-A10B-APEX-GGUF@Qwen3.5-122B-A10B-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",
    "base_model:Qwen/Qwen3.5-122B-A10B",
    "base_model:quantized:Qwen/Qwen3.5-122B-A10B",
    "license:apache-2.0",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 13,
  "downloads": 8550,
  "gated": false,
  "private": false,
  "last_modified": "2026-04-03T20:06:35.000Z",
  "created_at": "2026-04-02T14:46:53.000Z",
  "pipeline_tag": "",
  "library_name": ""
}

{
  "_id": "69ce815d6009007dc47ac747",
  "id": "mudler/Qwen3.5-122B-A10B-APEX-GGUF",
  "modelId": "mudler/Qwen3.5-122B-A10B-APEX-GGUF",
  "sha": "f1e7487c3bd3236297397b889c12c80463e6778e",
  "createdAt": "2026-04-02T14:46:53.000Z",
  "lastModified": "2026-04-03T20:06:35.000Z",
  "author": "mudler",
  "downloads": 8550,
  "likes": 13,
  "gated": false,
  "private": false,
  "pipeline_tag": "",
  "library_name": "",
  "siblings_count": 10
}