3 ms·
That sounds wild. While being ignorant about LLMs internals, I would have expected such things, crashes and session leaks, be impossible by design.
by febeling 3y ago
That sounds wild. While being ignorant about LLMs internals, I would have expected such things, crashes and session leaks, be impossible by design.
- shawntan 3y agoNaNs are not only possible by design, but are extremely common. Training of LLMs involve many tricks about how to deal with training steps that result in NaNs. Quantisation of LLMs also require dealing with huge outlier values.
- airgapstopgap 3y agoNote that we have no reason to believe that the underlying LLM inference process has suffered any setbacks. Obviously it has generated some logits. But the question is how is OpenAI server configured and what inference optimization tricks they're using.
- febeling 3y agoThe operation of this server is very uniform, in my imagination. Just emitting chunks of string. That this can be disrupted and an edge case occur, by the content of the strings - I find it puzzling.