4 ms·
There was no parameter creep with Llama. Llama 8B is actually a ~7B model comparable to Mistral 7B if you strip away multilingual embeddings and match what Mist
by kouteiheika 2y ago
There was no parameter creep with Llama. Llama 8B is actually a ~7B model comparable to Mistral 7B if you strip away multilingual embeddings and match what Mistral 7B supports.