Apr 30, 10:45 PM

Llama-3 Models Hit 1M Context on Hugging Face, New Releases by InternLM

The AI community is abuzz with significant advancements in Llama-3 models, as various new developments have been announced. Notably, the Llama-3-8B model now supports an unprecedented 1 million context length, a substantial increase from the previous 160K context length, and is available on Hugging Face. This enhancement was made possible through the sponsorship of Crusoe Energy. Additionally, the Llama-3 model, trained on 15 trillion tokens, has shown sensitivity to quantization degradation due to its high precision in BF16. Furthermore, new fine-tuned models such as Llama-3-70B and Llamixtral-3 are being released, alongside the introduction of Vision Language Models based on Llama-3 8B and Phi-3 Mini by the InternLM team.

#Llama #Hugging Face #Crusoe Energy #Vision Language Models #InternLM

Written with ChatGPT (GPT-4).