Meta AI released the latest generation of open-source large-scale model Llama 2 on Tuesday. Compared to Llama 1 released in February this year, Llama 2 doubled the number of tokens used for training, reaching 20 trillion, and also doubled the context length limit. Llama 2 includes models with 7 billion, 13 billion, and 70 billion parameters.
In addition, Meta also announced a collaboration with Microsoft Azure, the cloud service of Microsoft, to launch cloud services based on the Llama 2 model for global developers. This collaboration marks the deep cooperation between Meta and Microsoft, and officially launches the next generation of open-source large language model Llama 2, and announces that the model will be provided free of charge for research and commercial use. Compared to Llama 1, Llama 2 has increased its training data by 40%, the context length is twice that of Llama 1, and it adopts a grouped query attention mechanism.
Specifically, the pre-training model of Llama 2 is trained on 20 trillion tokens, and the fine-tuned Chat model is trained on 1 million human-labeled data.
Meta AI Llama 2 official introduction page: https://941ai.com/zyur
Llama 2 official paper address: https://941ai.com/dyog
Llama Github official open source code website: https://github.com/facebookresearch/llama
Committed to discovering the new world of artificial intelligence,
Continuously update the current popular AI tutorials and news!
Stay tuned to AI Exploration Station and explore the mysterious world of AI with us!
https://941ai.com/meta-llama2-1188.html