2 ms·
It looks like LLaMa is undertrained, especially 65B(that's why it looses on a couple of important benchmarks to 30B).
by terafo 4y ago
It looks like LLaMa is undertrained, especially 65B(that's why it looses on a couple of important benchmarks to 30B).