3 ms·
Why GTP-2? Seems like a weird choice unless it was made long ago. I believe chirper.ai is using Falcon now, as I saw they funded one if the finetunes.
by brucethemoose2 3y ago
Why GTP-2? Seems like a weird choice unless it was made long ago.
I believe chirper.ai is using Falcon now, as I saw they funded one if the finetunes.
- minimaxir 3y agoIt was made very long ago. (May 2019)
- behnamoh 3y agoIsn’t Falcon too slow at inference tho?
- brucethemoose2 3y agoThe GPTQ implementation is reasonably fast, I think. Or maybe they are running it at 8 bits? I just saw that they funded this: https://huggingface.co/ehartford/WizardLM-Uncensored-Falcon-40b https://huggingface.co/ehartford/WizardLM-Uncensored-Falcon-...
- behnamoh 3y agoInteresting! I don’t know what they’re up to and what’s the purpose of their project. But using an uncensored model is kinda risky in production, unless they know what they’re doing.