Nov 27, 05:11 PM

AI2 Launches OLMo 2 and SmolVLM, Advancing Open-Source Models with 7B, 13B Parameters, and 5T Tokens

The Allen Institute for AI (AI2) has released OLMo 2, a new family of open-source language models available in 7 billion and 13 billion parameters. These models have been trained on up to 5 trillion tokens, showcasing significant advancements in the field of artificial intelligence. In addition, the release of SmolVLM, a vision language model with 2.25 billion parameters, has been noted for its efficiency, requiring only 5GB of GPU RAM and being fine-tunable on Google Colab's free tier. Other developments in AI include the introduction of TULU 3, a fully open post-trained language model that outperforms several proprietary models, and the release of M-LongDoc, which aims to enhance AI's ability to process long documents. Furthermore, the latest version of Transformers.js has been launched, enabling advanced capabilities in browser-based AI applications. These innovations reflect the ongoing evolution of AI technologies, emphasizing both performance and accessibility.

#Allen Institute for AI #AI #SmolVLM #Google Colab

Written with ChatGPT (GPT-4o mini).

Sources

Additional media

Image #1 for story ai2-launches-olmo-2-smolvlm-advancing-open-source-models-7b-13b-parameters-5t-27926b9c

Image #2 for story ai2-launches-olmo-2-smolvlm-advancing-open-source-models-7b-13b-parameters-5t-27926b9c

Image #3 for story ai2-launches-olmo-2-smolvlm-advancing-open-source-models-7b-13b-parameters-5t-27926b9c

Image #4 for story ai2-launches-olmo-2-smolvlm-advancing-open-source-models-7b-13b-parameters-5t-27926b9c