4 ms·
Just my two cents from experience, any sufficiently advanced LLM training or inference pipeline eventually figures out that the real bottleneck is the network!
by moralestapia 2y ago
Just my two cents from experience, any sufficiently advanced LLM training or inference pipeline eventually figures out that the real bottleneck is the network!