3 ms·
This is not the case for LLMs. FP16/BF16 training precision is standard, with FP8 inference very common. But labs are moving to FP8 training and even FP4.
by qeternity 11mo ago
This is not the case for LLMs. FP16/BF16 training precision is standard, with FP8 inference very common. But labs are moving to FP8 training and even FP4.