

Recent developments in AI models have seen significant advancements with the release of various versions of Llama-3 models. These models, ranging from 8B to 70B, have been fine-tuned and optimized for improved performance. The context length of these models has been a key focus, with efforts to increase it beyond the initial limitations. The open-source LLama-3 8B model with a context length of over 1M has been highlighted as a game-changer for local AI on devices, showcasing impressive testing results and optimization efforts.
In ~2 months we went from Gemini-1.5 dominating 1M context use cases but very limited access to a 1M context adaptor anyone can attach to their llama3 finetune. Unbelievable progress, enabled by some unreal engineers at @Gradient_AI_ and open-source champion @winglian What's… https://t.co/F0QDDrzRpz
𝐋𝐋𝐚𝐕𝐀-𝐋𝐥𝐚𝐦𝐚-𝟑-𝟖𝐁 model 🚀 by @intern_lm - A LLaVA model fine-tuned from Meta-Llama-3-8B-Instruct & CLIP-ViT-Large-patch14-336 🥳LLaVA-Phi-3 Mini is available too. It outperforms LLaVA-v1.5-7B and matches the performance of LLaVA-Llama-3-8B 🤯 in multiple benchmarks. https://t.co/ZLfGIwZX0N
By far, the best way to run the new 1048k context Llama 3 8B IT fine tune with a 2k context length. 🤣 ps: Meta-Llama-3-8B-Instruct supports 8k context. https://t.co/Okf2Rqj0jP https://t.co/7Kv3nai30p