Jan 13, 06:06 PM

Kyutai Labs Launches Helium-1 Preview, 2B Parameter Multilingual LLM for Mobile Devices, Trained on 2.5 Trillion Tokens

Kyutai Labs has launched the Helium-1 Preview, a 2 billion parameter multilingual large language model (LLM) designed for edge and mobile devices. The model, which is released under a CC-BY-4.0 license, has been trained on 2.5 trillion tokens and features a 4096 context size. It is reported to outperform or be comparable to other models such as Qwen 2.5 1.5B, Owen 1.5B, Gemma 2B, and Llama 3B. The Helium-1 model is now available for developers to utilize on the Hugging Face platform, with plans for stronger text models and multimodal capabilities in the future.

#Kyutai Labs #Qwen #Owen #Gemma 2B #Llama 3B #Hugging Face

Written with ChatGPT (GPT-4o mini).

Sources

Jeff Boudier 🤗@jeffboudier
2 years ago
.@kyutai_labs just dropped Helium-1 Preview on @huggingface 📲 2B Base LLM for mobile / edge 🌍 Multi-lingual: 🇬🇧🇫🇷🇩🇪🇪🇸🇮🇹🇵🇹 ⚖️ CC-BY-4.0 license 🚀 Available today in transformers (pull main) Helium-1 Preview shows strong capabilities in multi-modal, and a full model is in the… https://t.co/DQzgZGREi4
Alexandre Défossez@honualx
2 years ago
We just released the Helium-1 model , a 2B multi-lingual LLM which @EXGRV and @lmazare have been crafting for us! Best model so far under 2.17B params on multi-lingual benchmarks 🇬🇧🇮🇹🇪🇸🇵🇹🇫🇷🇩🇪 On HF, under CC-BY licence: https://t.co/zpEo6dmJXV https://t.co/9GtZdzwzqV https://t.co/p5ZyZsICFO
Neil Zeghidour@neilzegh
2 years ago
Very proud to see our first standalone text LLM released. In our multimodal roadmap, we plan on releasing stronger text models along with audio, vision and more! https://t.co/Ez2JZlWNOz

Kyutai Labs Launches Helium-1 Preview, 2B Parameter Multilingual LLM for Mobile Devices, Trained on 2.5 Trillion Tokens

Sources

Additional media