| --- |
| title: README |
| emoji: 🌏 |
| colorFrom: blue |
| colorTo: indigo |
| sdk: static |
| pinned: false |
| --- |
| |
| <p align="center"> |
| <img src="https://cdn-avatars.huggingface.co/v1/production/uploads/6547ee216b1695ac33d76867/PRjMaJNZfh53VTOXlqzDo.png" width="88" alt="GRINDA AI"/> |
| </p> |
|
|
| # GRINDA AI |
|
|
| **Applied AI research from Korea, shipped into real products.** |
| GRINDA AI builds AI agents for global B2B. [Rinda](https://www.rinda.ai/en) finds overseas buyers and writes outreach in their own language; [Ara](https://www.grinda.ai/en/products/ara) is an AI employee that works inside your team's messenger. The models behind them start here. |
|
|
| Our open models and datasets are published under **[FINGU-AI](https://huggingface.co/FINGU-AI)**: 100+ models and 30+ datasets. |
|
|
| ## Research areas |
|
|
| | Area | What we release | Browse | |
| |---|---|---| |
| | **Low-resource multilingual translation** | Qwen2.5 7B / 32B translation LLMs for Korean and Uzbek, Nepali, Khmer, Burmese, Mongolian, Sinhala, Tagalog, Thai, Vietnamese and more, plus scored evaluation sets | [Collection](https://huggingface.co/collections/FINGU-AI/low-resource-multilingual-translation-6af163703692a6fca0914b36) | |
| | **Embeddings & retrieval** | FingU sentence-embedding models (gte-Qwen2, stella and mpnet bases) evaluated on MTEB, with retrieval, reranking and STS data | [Collection](https://huggingface.co/collections/FINGU-AI/fingu-embeddings-6af1635943c4b34a0ae6c731) | |
| | **Finance-domain LLMs** | English / Korean / Japanese finance chat models and translated finance corpora | [Collection](https://huggingface.co/collections/FINGU-AI/finance-llms-and-data-6af163739ba5237bdad2bf08) | |
| | **Model merging** | SLERP and linear merges at 14B-32B for reasoning and multi-turn chat | [Collection](https://huggingface.co/collections/FINGU-AI/model-merges-14b-32b-6af16375a80d7c3ba0204050) | |
|
|
| ## Start with these |
|
|
| - [**QWEN2.5-32B-2600s**](https://huggingface.co/FINGU-AI/QWEN2.5-32B-2600s): Qwen2.5-32B fine-tuned for translation across 16 languages ([FP8](https://huggingface.co/FINGU-AI/QWEN2.5-32B-2600s-FP8)) |
| - [**FingUEm_V3**](https://huggingface.co/FINGU-AI/FingUEm_V3): 1.5B embedding model with full MTEB results in the card |
| - [**Chocolatine-Fusion-14B**](https://huggingface.co/FINGU-AI/Chocolatine-Fusion-14B): our most-liked merge |
| - [**FinguAI-Chat-v1**](https://huggingface.co/FINGU-AI/FinguAI-Chat-v1): finance chat model for English, Korean and Japanese |
|
|
| Licenses vary by model; check each model card. |
|
|
| ## About us |
|
|
| - Founded in 2023 in Daejeon, South Korea; selected for Korea's TIPS deep-tech startup program |
| - Exhibited at CES 2024; building AI with finance, government and global partners |
| - Products: [Rinda](https://www.rinda.ai/en) (AI sales agent), [Ara](https://www.grinda.ai/en/products/ara) (AI employee), AI document OCR |
|
|
| ## Work with us |
|
|
| Working on multilingual or low-resource-language AI, retrieval, or agents for real business workflows? We welcome research collaborations, data partnerships and feedback on our models. |
|
|
| - Questions about a model: open a discussion on its page |
| - Partnerships: [grinda.ai/en/contact](https://www.grinda.ai/en/contact) |
| - [Website](https://www.grinda.ai/en) · [GitHub](https://github.com/FINGU-GRINDA) · [LinkedIn](https://www.linkedin.com/company/grinda-ai) |
|
|
