2 ms·
Architectural changes between Llama 2 and 3 seem to be minimal. Looking at the 400B model benchmarks and comparing them to GPT-4 only proves that there is no se
by lossolo 3y ago
Architectural changes between Llama 2 and 3 seem to be minimal. Looking at the 400B model benchmarks and comparing them to GPT-4 only proves that there is no secret sauce. It's all about the dataset and the number of params.