lgai-exaone/exaone-deep-32b-gguf overview

{
  "metadata": {},
  "card_data": {
    "base_model": "LGAI-EXAONE/EXAONE-Deep-32B",
    "base_model_relation": "quantized",
    "license": "other",
    "license_name": "exaone",
    "license_link": "LICENSE",
    "language": [
      "en",
      "ko"
    ],
    "tags": [
      "lg-ai",
      "exaone",
      "exaone-deep"
    ],
    "pipeline_tag": "text-generation",
    "library_name": "transformers",
    "frontmatter": {
      "base_model": "LGAI-EXAONE/EXAONE-Deep-32B",
      "base_model_relation": "quantized",
      "license": "other",
      "license_name": "exaone",
      "license_link": "LICENSE",
      "language": [
        "en",
        "ko"
      ],
      "tags": [
        "lg-ai",
        "exaone",
        "exaone-deep"
      ],
      "pipeline_tag": "text-generation",
      "library_name": "transformers"
    },
    "hero_image_url": "assets/EXAONE_Symbol+BI_3d.png",
    "summary": "",
    "quick_links": [],
    "benchmark_table_html": "",
    "readme_markdown": "---\nbase_model: LGAI-EXAONE/EXAONE-Deep-32B\nbase_model_relation: quantized\nlicense: other\nlicense_name: exaone\nlicense_link: LICENSE\nlanguage:\n- en\n- ko\ntags:\n- lg-ai\n- exaone\n- exaone-deep\npipeline_tag: text-generation\nlibrary_name: transformers\n---\n\n<p align=\"center\">\n<img src=\"assets/EXAONE_Symbol+BI_3d.png\", width=\"300\", style=\"margin: 40 auto;\">\n<br>\n\n# EXAONE-Deep-32B-GGUF\n\n## Introduction\n\nWe introduce EXAONE Deep, which exhibits superior capabilities in various reasoning tasks including math and coding benchmarks, ranging from 2.4B to 32B parameters developed and released by LG AI Research. Evaluation results show that 1) EXAONE Deep **2.4B** outperforms other models of comparable size, 2) EXAONE Deep **7.8B** outperforms not only open-weight models of comparable scale but also a proprietary reasoning model OpenAI o1-mini, and 3) EXAONE Deep **32B** demonstrates competitive performance against leading open-weight models.\n\nFor more details, please refer to our [documentation](https://arxiv.org/abs/2503.12524), [blog](https://www.lgresearch.ai/news/view?seq=543) and [GitHub](https://github.com/LG-AI-EXAONE/EXAONE-Deep).\n\n<p align=\"center\">\n<img src=\"assets/exaone_deep_overall_performance.png\", width=\"100%\", style=\"margin: 40 auto;\">\n\nThis repository contains the various precisions of the reasoning 32B language model in GGUF format, which contains the following features:\n\n- Number of Parameters (without embeddings): 30.95B\n- Number of Layers: 64\n- Number of Attention Heads: GQA with 40 Q-heads and 8 KV-heads\n- Vocab Size: 102,400\n- Context Length: 32,768 tokens\n- Quantization: `Q8_0`, `Q6_K`, `Q5_K_M`, `Q4_K_M`, `IQ4_XS` in GGUF format (also includes `BF16` weights)\n\n## Quickstart\n\nHere are the steps to run conversational inference with the model:\n\n1. Install llama.cpp. Please refer to the [llama.cpp repository](https://github.com/ggerganov/llama.cpp) for more details.\n\n2. Download EXAONE Deep model in GGUF format.\n\n```bash\nhuggingface-cli download LGAI-EXAONE/EXAONE-Deep-32B-GGUF \\\n    --include \"EXAONE-Deep-32B-BF16*.gguf\" \\\n    --local-dir .\n```\n\n3. (Optional) In case of using BF16 precision, you may need to merge the split files.\n```bash\nllama-gguf-split --merge \\\n    ./EXAONE-Deep-32B-BF16-00001-of-00002.gguf \\\n    ./EXAONE-Deep-32B-BF16.gguf\n```\n\n4. Run the model with llama.cpp in conversational mode. We set chat-template explicitly to handle reasoning steps properly. \n\n```bash\nllama-cli -m ./EXAONE-Deep-32B-BF16.gguf \\\n    -sys \"\" \\\n    -c 32768 \\\n    --temp 0.6 \\\n    --top-p 0.95 \\\n    --jinja \\\n    --chat-template \"{% for message in messages %}{% if loop.first and message['role'] != 'system' %}{{ '[|system|][|endofturn|]\\n' }}{% endif %}{% set content = message['content'] %}{% if '</thought>' in content %}{% set content = content.split('</thought>')[-1].lstrip('\\\\n') %}{% endif %}{{ '[|' + message['role'] + '|]' + content }}{% if not message['role'] == 'user' %}{{ '[|endofturn|]' }}{% endif %}{% if not loop.last %}{{ '\\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '\\n[|assistant|]<thought>\\n' }}{% endif %}\"\n```\n\n> ### Note\n> The EXAONE Deep models are trained with an optimized configuration,\n> so we recommend following the [Usage Guideline](#usage-guideline) section to achieve optimal performance.\n\n## Evaluation\n\nYou can check the evaluation results of original EXAONE Deep models at [GitHub](https://github.com/LG-AI-EXAONE/EXAONE-Deep) or our [documentation](https://arxiv.org/abs/2503.12524).\n\n## Deployment\n\nEXAONE Deep models can be inferred in the various frameworks, such as:\n- `TensorRT-LLM`\n- `vLLM`\n- `SGLang`\n- `llama.cpp`\n- `Ollama`\n- `LM-Studio`\n\nPlease refer to our [EXAONE Deep GitHub](https://github.com/LG-AI-EXAONE/EXAONE-Deep) for more details about the inference frameworks.\n\n## Quantization\n\nWe provide the pre-quantized EXAONE Deep models with **AWQ** and several quantization types in **GGUF** format. Please refer to our [EXAONE Deep collection](https://huggingface.co/collections/LGAI-EXAONE/exaone-deep-67d119918816ec6efa79a4aa) to find corresponding quantized models.\n\n## Usage Guideline\n\nTo achieve the expected performance, we recommend using the following configurations:\n\n1. Ensure the model starts with `<thought>\\n` for reasoning steps. The model's output quality may be degraded when you omit it. You can easily apply this feature by using `tokenizer.apply_chat_template()` with `add_generation_prompt=True`. Please check the example code on [Quickstart](#quickstart) section.\n2. The reasoning steps of EXAONE Deep models enclosed by `<thought>\\n...\\n</thought>` usually have lots of tokens, so previous reasoning steps may be necessary to be removed in multi-turn situation. The provided tokenizer handles this automatically.\n3. Avoid using system prompt, and build the instruction on the user prompt. \n4. Additional instructions help the models reason more deeply, so that the models generate better output.\n    - For math problems, the instructions **\"Please reason step by step, and put your final answer within \\boxed{}.\"** are helpful.\n    - For more information on our evaluation setting including prompts, please refer to our [Documentation](https://arxiv.org/abs/2503.12524).\n5. In our evaluation, we use `temperature=0.6` and `top_p=0.95` for generation. \n6. When evaluating the models, it is recommended to test multiple times to assess the expected performance accurately.\n\n## Limitation\n\nThe EXAONE language model has certain limitations and may occasionally generate inappropriate responses. The language model generates responses based on the output probability of tokens, and it is determined during learning from training data. While we have made every effort to exclude personal, harmful, and biased information from the training data, some problematic content may still be included, potentially leading to undesirable responses. Please note that the text generated by EXAONE language model does not reflects the views of LG AI Research.\n\n- Inappropriate answers may be generated, which contain personal, harmful or other inappropriate information.\n- Biased responses may be generated, which are associated with age, gender, race, and so on.\n- The generated responses rely heavily on statistics from the training data, which can result in the generation of\nsemantically or syntactically incorrect sentences.\n- Since the model does not reflect the latest information, the responses may be false or contradictory.\n\nLG AI Research strives to reduce potential risks that may arise from EXAONE language models. Users are not allowed\nto engage in any malicious activities (e.g., keying in illegal information) that may induce the creation of inappropriate\noutputs violating LG AI’s ethical principles when using EXAONE language models.\n\n## License\n\nThe model is licensed under [EXAONE AI Model License Agreement 1.1 - NC](./LICENSE)\n\n## Citation\n \n```\n@article{exaone-deep,\n  title={EXAONE Deep: Reasoning Enhanced Language Models},\n  author={{LG AI Research}},\n  journal={arXiv preprint arXiv:2503.12524},\n  year={2025}\n}\n```\n\n## Contact\nLG AI Research Technical Support: contact_us@lgresearch.ai",
    "related_quantizations": []
  },
  "tags": [
    "transformers",
    "gguf",
    "lg-ai",
    "exaone",
    "exaone-deep",
    "text-generation",
    "en",
    "ko",
    "arxiv:2503.12524",
    "base_model:LGAI-EXAONE/EXAONE-Deep-32B",
    "base_model:quantized:LGAI-EXAONE/EXAONE-Deep-32B",
    "license:other",
    "endpoints_compatible",
    "region:us",
    "conversational"
  ],
  "likes": 46,
  "downloads": 555,
  "gated": false,
  "private": false,
  "last_modified": "2025-03-18T12:49:59.000Z",
  "created_at": "2025-03-12T04:44:23.000Z",
  "pipeline_tag": "text-generation",
  "library_name": "transformers"
}

{
  "_id": "67d111272187e07c9e33633e",
  "id": "LGAI-EXAONE/EXAONE-Deep-32B-GGUF",
  "modelId": "LGAI-EXAONE/EXAONE-Deep-32B-GGUF",
  "sha": "4ba3a5454685aa7fda2b77c62c810879c084d78c",
  "createdAt": "2025-03-12T04:44:23.000Z",
  "lastModified": "2025-03-18T12:49:59.000Z",
  "author": "LGAI-EXAONE",
  "downloads": 555,
  "likes": 46,
  "gated": false,
  "private": false,
  "pipeline_tag": "text-generation",
  "library_name": "transformers",
  "siblings_count": 12
}