Llama 3.3 has been released! https://huggingface.co/meta-llama/Llama-3.3-70B-Instruct The 70B model has been fine-tuned to the point where it occasionally outperforms the 405B model. There's a particularly significant improvement in math and coding tasks, where Llama has traditionally been weaker. This time, only the 70B model is being released—there are no other sizes or VLM versions.
[link] [comments]