4 ms·
Check out this post from a round table dialogue with Greg Brockman from OpenAI. The GPT models that were in existence / in use in early 2023 were not the perfor
by hungrigekatze 3y ago
Check out this post from a round table dialogue with Greg Brockman from OpenAI. The GPT models that were in existence / in use in early 2023 were not the performance-degraded quantized versions that are in production now: https://www.reddit.com/r/mlscaling/comments/146rgq2/chatgpt_is_running_quantized/jnro7wo/ https://www.reddit.com/r/mlscaling/comments/146rgq2/chatgpt_...
- sroussey 3y agoOh interesting. I thought that’s what turbo was.
- refulgentis 3y agoIt was, that's what the comment says?