\n\n![](\"assets/EXAONE_Symbol+BI_3d.png\",)
\n
\n
🤗 HuggingFace | 📝 Blog | 📑 Technical Report \n
\n\n
\n\n## Introduction\n\nWe introduce EXAONE-3.0-7.8B-Instruct, a pre-trained and instruction-tuned bilingual (English and Korean) generative model with 7.8 billion parameters. \nThe model was pre-trained with 8T curated tokens and post-trained with supervised fine-tuning and direct preference optimization. \nIt demonstrates highly competitive benchmark performance against other state-of-the-art open models of similar size. \n\n
\n\n## News\n\n- 2024.08.08: :wave: :wave: We have revised our [license](./LICENSE) for revitalizing the research ecosystem. :wave: :wave:\n- 2024.08.07: We released the EXAONE 3.0 7.8B instruction-tuned model. Check out the 📑 [Technical Report](https://arxiv.org/abs/2408.03541)!\n\n
\n\n## Performance\n\nSome experimental results are shown below. The full evaluation results can be found in the [Technical Report](https://arxiv.org/abs/2408.03541).\n\n| Language | Benchmark | EXAONE 3.0
7.8B Inst. | Llama 3.1
8B Inst. | Gemma 2
9B Inst. | QWEN 2
7B Inst. | Phi 3
7B Inst. | Mistral 7B
Inst. |\n| :-----: | :----- | :-----: | :-----: | :-----: | :-----: | :-----: | :-----: |\n| English | MT-Bench | **9.01** | 7.95 | 8.52 | 8.41 | 8.52 | 7.72 |\n| | Arena-Hard-v0.1 | **46.8** | 28.0 | 42.1 | 21.7 | 29.1 | 16.2 |\n| | WildBench | **48.2** | 34.5 | 41.5 | 34.9 | 32.8 | 29.0 |\n| | AlpacaEval 2.0 LC | 45.0 | 31.5 | **47.5** | 24.5 | 37.1 | 31.0 |\n| Korean | KoMT-Bench[^1] | **8.92** | 6.06 | 7.92 | 7.69 | 4.87 | 5.20 |\n| | LogicKor | **8.62** | 5.40 | 8.07 | 6.12 | 3.76 | 3.42 |\n\n
\n\n## Requirements\n"},{"ref":"P20","kind":"page","title":"LG-AI-EXAONE/EXAONEPath repository metadata","date":"2026-06-11T03:22:48.248562+00:00","date_source":null,"source_url":"https://github.com/LG-AI-EXAONE/EXAONEPath","signal_url":null,"signal_json_url":null,"text":"# LG-AI-EXAONE/EXAONEPath\n\nLanguage: Python\n\nLicense: NOASSERTION\n\nStars: 23\n\nForks: 3\n\nOpen issues: 1\n\nCreated: 2024-07-31T04:53:46Z\n\nPushed: 2025-01-08T07:08:34Z\n\nDefault branch: main\n\nFork: no\n\nArchived: no\n\nREADME:\n\n\n# EXAONEPath\n\n## EXAONEPath 1.0 Patch-level Foundation Model for Pathology\n\n[[`Paper`](https://arxiv.org/abs/2408.00380)] [[`Hugging Face`](https://huggingface.co/LGAI-EXAONE/EXAONEPath)] [[`Model`](https://github.com/LG-AI-EXAONE/EXAONEPath/releases/download/1.0.0/EXAONEPath.ckpt)] [[`BibTeX`](#Citation)]\n\n\n\n\n## Introduction\nWe introduce EXAONEPath, a patch-level pathology pretrained model with 86 million parameters. \nThe model was pretrained on 285,153,903 patches extracted from a total of 34,795 WSIs. \nEXAONEPath demonstrates superior performance considering the number of WSIs used and the model's parameter count.\n\n## Quickstart\nLoad EXAONEPath and run inference on tile-level images.\n\n### 1. Hardware Requirements ###\n- NVIDIA GPU is required\n- Minimum 8GB GPU memory recommended\n- NVIDIA driver version >= 450.80.02 required\n\nNote: This implementation requires NVIDIA GPU and drivers. The provided environment setup specifically uses CUDA-enabled PyTorch, making NVIDIA GPU mandatory for running the model.\n\n### 2. Environment Setup ###\nFirst, install Conda if you haven't already. You can find installation instructions [here](https://docs.anaconda.com/miniconda/).\nThen create and activate the environment using the provided configuration:\n```bash\ngit clone https://github.com/LG-AI-EXAONE/EXAONEPath.git\ncd EXAONEPath\nconda env create -f environment.yaml\nconda activate exaonepath\n```\n\n### 3. Load the model & Inference\n#### Load with HuggingFace\n\n```python\nimport torch\nfrom PIL import Image\nfrom macenko import macenko_normalizer\nimport torchvision.transforms as transforms\nfrom vision_transformer import VisionTransformer\n\nhf_token = \"YOUR_HUGGING_FACE_ACCESS_TOKEN\"\nmodel = VisionTransformer.from_pretrained(\"LGAI-EXAONE/EXAONEPath\", use_auth_token=hf_token)\n\ntransform = transforms.Compose(\n[\ntransforms.Resize(256"},{"ref":"P21","kind":"page","title":"LG-AI-EXAONE/EXAONE-Examples repository metadata","date":"2026-06-11T03:22:48.245375+00:00","date_source":null,"source_url":"https://github.com/LG-AI-EXAONE/EXAONE-Examples","signal_url":null,"signal_json_url":null,"text":"# LG-AI-EXAONE/EXAONE-Examples\n\nLanguage: Jupyter Notebook\n\nStars: 1\n\nForks: 0\n\nOpen issues: 0\n\nCreated: 2024-09-20T05:56:03Z\n\nPushed: 2024-10-02T05:18:13Z\n\nDefault branch: main\n\nFork: no\n\nArchived: no\n\nREADME:\n# EXAONE Examples\n\nJupyter notebook examples for EXAONE Text Generation in AWS Marketplace."},{"ref":"P22","kind":"page","title":"LG-AI-EXAONE/EXAONE-4.0 repository metadata","date":"2026-06-11T03:22:46.958804+00:00","date_source":null,"source_url":"https://github.com/LG-AI-EXAONE/EXAONE-4.0","signal_url":null,"signal_json_url":null,"text":"# LG-AI-EXAONE/EXAONE-4.0\n\nDescription: Official repository for EXAONE 4.0 built by LG AI Research\n\nLicense: NOASSERTION\n\nStars: 105\n\nForks: 9\n\nOpen issues: 6\n\nCreated: 2025-07-07T06:45:57Z\n\nPushed: 2025-08-04T02:19:34Z\n\nDefault branch: main\n\nFork: no\n\nArchived: no\n\nREADME:\n# EXAONE-4.0\n
\n
\n
\n\n
\n\n
\n
\n\n
\n\n
\n\n| 🎉 License Updated! We are pleased to announce our more flexible licensing terms 🤗 [What's Different?](#license) |\n|:---:|\n\n
\n\n## Introduction\n\nWe introduce **EXAONE 4.0**, which integrates a **Non-reasoning mode** and **Reasoning mode** to achieve both the excellent usability of [EXAONE 3.5](https://github.com/LG-AI-EXAONE/EXAONE-3.5) and the advanced reasoning abilities of [EXAONE Deep](https://github.com/LG-AI-EXAONE/EXAONE-Deep). To pave the way for the agentic AI era, EXAONE 4.0 incorporates essential features such as agentic tool use, and its multilingual capabilities are"},{"ref":"P23","kind":"page","title":"LG-AI-EXAONE/EXAONE-Deep repository metadata","date":"2026-06-11T03:22:46.952321+00:00","date_source":null,"source_url":"https://github.com/LG-AI-EXAONE/EXAONE-Deep","signal_url":null,"signal_json_url":null,"text":"# LG-AI-EXAONE/EXAONE-Deep\n\nDescription: Official repository for EXAONE Deep built by LG AI Research\n\nLicense: NOASSERTION\n\nStars: 401\n\nForks: 27\n\nOpen issues: 6\n\nCreated: 2025-03-12T06:45:27Z\n\nPushed: 2025-06-02T00:22:33Z\n\nDefault branch: main\n\nFork: no\n\nArchived: no\n\nREADME:\n# EXAONE Deep\n
\n\n\n
\n
🤗 Hugging Face | 📝 Blog | 📑 Documentation \n
\n\n
\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\n
\n
\n![](\"assets/exaone_deep_overall_performance.png\",)
\n
\n\nOur documentation consists of the following sections:\n\n- [Performance](#performance): Experimental results of EXAONE Deep models.\n- [Quickstart](#quickstart): A basic guide to using EXAONE Deep models with Transformers.\n- [Quantized Models](#quantized-models): An explanation of quantized EXAONE Deep weights in `AWQ` and `GGUF` format.\n- [Run Locally](#run-locally): A guide to running EXAONE Deep models locally with `llama.cpp` and `Ollama` frameworks.\n- [Deployment](#deployment): A guide to running EXAONE Deep models with `TensorRT-LLM`, `vLLM`, and `SGLang` deployment frameworks.\n- [Usage Guideline](#usage-guideline): A guide to utilizing EXAONE Deep models to achieve the expected performance.\n\n
\n\n## News\n\n- 2025.03.18: We release the EXAONE Deep, reasoning enhanced language models, including 2.4B, 7.8B"},{"ref":"P24","kind":"page","title":"LG-AI-EXAONE/KMMLU-Pro repository metadata","date":"2026-06-11T03:22:46.939399+00:00","date_source":null,"source_url":"https://github.com/LG-AI-EXAONE/KMMLU-Pro","signal_url":null,"signal_json_url":null,"text":"# LG-AI-EXAONE/KMMLU-Pro\n\nLanguage: Python\n\nLicense: BSD-3-Clause\n\nStars: 16\n\nForks: 1\n\nOpen issues: 0\n\nCreated: 2025-08-14T05:29:37Z\n\nPushed: 2025-08-18T05:57:30Z\n\nDefault branch: main\n\nFork: no\n\nArchived: no\n\nREADME:\n# KMMLU-Pro Evaluation Script\n\n**Language**: [English](README.md) | [한국어](README_ko.md)\n\n📄 [Paper](https://arxiv.org/abs/2507.08924) | 📚 [Dataset](https://huggingface.co/datasets/LGAI-EXAONE/KMMLU-Pro)\n\n## Overview\n\n**KMMLU-Pro** is a challenging benchmark comprising 2,822 problems from the 2024 Korean National Professional Licensure (KNPL) official exams, representing highly specialized professions in Korea.\nThis repository provides evaluation scripts to generate model responses using the OpenAI-compatible interface and calculate professional license pass/fail results.\n\n## Setup\n\n### Prerequisites\n\n1. **Dataset Access**: Request access to the [KMMLU-Pro dataset](https://huggingface.co/datasets/LGAI-EXAONE/KMMLU-Pro) on Hugging Face.\n\n2. **OpenAI API Key**: Set your OpenAI API key as an environment variable:\n```bash\nexport OPENAI_API_KEY=\"your-api-key-here\"\n```\n\n3. **Installation**:\n```bash\ngit clone https://github.com/LG-AI-EXAONE/KMMLU-Pro.git\ncd KMMLU-Pro\npip install -r requirements.txt\n```\n\n## Usage\n\n### 1. Generate Model Responses\n\n**Non-Reasoning Model Usage**\n```bash\npython generate_model_responses.py --model YOUR_MODEL_NAME\n```\n\n**Reasoning Model Usage**\n```bash\npython generate_model_responses.py --model YOUR_MODEL_NAME --temperature 0.6 --top_p 0.95 --enable_reasoning\n```\n\n**Additional Options**\n- `--model`: Model name (required)\n- `--output_dir`: Output directory (default: `./results`)\n- `--prompt_language`: Prompt language, 'ko' or 'en' (default: `ko`)\n- `--temperature`: Sampling temperature (default: `0.0`)\n- `--top_p`: Top-p sampling (default: `1.0`)\n- `--presence_penalty`: Presence penalty (default: `0.0`)\n- `--max_tokens`: Maximum tokens per response (default: `32768`)\n- `--max_requests`: Maximum concurrent requests (default: `200`)\n- `--enable_reasoning`: Enable reasoning mode (flag)\n\n### 2. Calculate Scores and License Results\n\n```bash\npython print_score.py --model_responses \"results/{YOUR_MODEL_NAME}_results.jsonl\"\n```\n\n## Outp"},{"ref":"P25","kind":"page","title":"LG-AI-EXAONE/EXAONE-Path-2.5 repository metadata","date":"2026-06-11T03:22:46.89067+00:00","date_source":null,"source_url":"https://github.com/LG-AI-EXAONE/EXAONE-Path-2.5","signal_url":null,"signal_json_url":null,"text":"# LG-AI-EXAONE/EXAONE-Path-2.5\n\nLanguage: Python\n\nLicense: NOASSERTION\n\nStars: 5\n\nForks: 0\n\nOpen issues: 0\n\nCreated: 2025-12-15T09:37:18Z\n\nPushed: 2026-03-10T01:53:23Z\n\nDefault branch: main\n\nFork: no\n\nArchived: no\n\nREADME:\n\n\n# EXAONE Path 2.5\n[[`Github`](https://github.com/LG-AI-EXAONE/EXAONE-Path-2.5)] [[`Hugging Face`](https://huggingface.co/LGAI-EXAONE/EXAONE-Path-2.5)] [[`Paper`](https://arxiv.org/abs/2512.14019)] [[`Cite`](#citation)]\n\n## Introduction\nEXAONE Path 2.5 is a biologically informed multimodal framework that enriches histopathology representations by aligning whole-slide images with *genomic, epigenetic, and transcriptomic data*. By enabling all-pairwise cross-modal alignment across multiple layers of tumor biology, the model captures coherent genotype-to-phenotype relationships within a unified embedding space. This domain-informed design improves resource efficiency, enabling the model to achieve competitive performance across diverse tasks while using substantially fewer training samples and parameters than existing approaches.\n\n
\n
![](\"https://raw.githubusercontent.com/LG-AI-EXAONE/EXAONE-Path-2.5/main/figs/exaonepath_2.5.png\")
\n
Figure 1. Overall scheme of EXAONE Path 2.5.\n
![\"Huggingface\"](\"https://img.shields.io/badge/🤗-Huggingface-FC926C?style=for-the-badge\"){kind=icon}
\n![\"Blog\"](\"https://img.shields.io/badge/📝-Blog-E343BD?style=for-the-badge\"){kind=icon}
\n![\"Technical](\"https://img.shields.io/badge/📑-Technical_Report-684CF4?style=for-the-badge\"){kind=icon}
\n![\"FriendliAI\"](\"https://img.shields.io/badge/✈️_commercial_purposes-Try_on_FriendliAI-2649BC?style=for-the-badge\"){kind=icon}
\n