Loading...
According to tweets, a paper by Microsoft Research titled 'CodeFusion: A Pre-trained Diffusion Model for Code Generation' suggests that OpenAI's GPT-3.5 Turbo has only 20 billion parameters. This leaked information is considered impressive given the quality of chat achieved with such a parameter count.
New Microsoft codediffusion paper suggests GPT-3.5 Turbo is only 20B https://t.co/Or5ys4Tqm2
Microsoft might've just leaked the "secretive" parameter count of GPT 3.5 in their CodeFusion paper. Achieving this quality of chat with 20B parameter is very impressive. https://t.co/AarJCZsiT0 https://t.co/TDQQJjcmIj
This is quite interesting - ChatGPT-3.5-turbo is just 20B params ?? This is from a paper by people from Microsoft Research "CODEFUSION: A Pre-trained Diffusion Model for Code Generation " https://t.co/YAb2GJOHEi https://t.co/DBpPFEC9zP https://t.co/nEpSMXxZHT
Did @MSFTResearch leak the parameter count of @OpenAI GPT-3.5 turbo?🤯 According to „CodeFusion: A Pre-trained Diffusion Model for Code Generation“ paper gpt-3.5-turbo has only 20B parameter. Paper: https://t.co/5ehQLffJWY https://t.co/v0MFxGLfaI
GPT-4 possibly trained on 25k A100 GPUs... GPT-5 if trained on the same # of H100s could scale to be 2.5x bigger than GPT-4 using fp8 https://t.co/FtN2VSnDMe