3 ms·

I forgot to clarify dealing with the network bottleneck

• moralestapia 1y 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!