Hacker News

Machine Learning News

TechCrunch

<p>AI21 Labs has recently unveiled Jamba, a ground-breaking hybrid SSM-Transformer large language model. Jamba combines the best attributes of the Mamba and Transformer models, aiming to address the limitations of transformers in generative reasoning tasks. With its hybrid architecture and optimized memory, Jamba outperforms existing models in terms of memory efficiency, context handling, and throughput. One of the key advantages of Jamba is its ability to handle larger context windows, making it a highly versatile model. It can run on a single GPU with at least 80GB of memory and offers impressive performance and efficiency gains. While Jamba may not entirely replace transformer-based models, it demonstrates superior capabilities in handling complex generative AI tasks. AI21 Labs plans to fine-tune Jamba for commercial use in the future, solidifying its position as the first production-grade AI model based on the Mamba architecture.</p>


Collections

AI21 Labs introduces Jamba, a hybrid SSM-Transformer model that outperforms existing models in generative reasoning tasks. It has optimized memory, improved context handling, and impressive performance on a single GPU with at least 80GB of memory. Jamba shows superior capabilities in handling complex generative AI tasks.

AI21 Labs Unveils Jamba: A State-of-the-Art AI Model with Enhanced Context Handling