4 ms·
Pythia was trained on only 300B tokens and is pretty dumb compared to LLaMA. Pythia 13B is worse than LLaMA-7B and requires double the resources.
by MacsHeadroom 4y ago
Pythia was trained on only 300B tokens and is pretty dumb compared to LLaMA.
Pythia 13B is worse than LLaMA-7B and requires double the resources.
- Tepix 4y agoNot all use cases need GPT-4 level performance. I'd argue that even LLaMA-7B is quite limited. Also, new and improved models are being released all the time.