Flan-PaLM 540B (Google) and GPT-4 (Jun 2023) (OpenAI) are both frontier AI models. Flan-PaLM 540B was published in October 2022 and GPT-4 (Jun 2023) in June 2023.
GPT-4 (Jun 2023) was trained on 2.1 x 10^25 FLOP, about 8.3x the compute of Flan-PaLM 540B at 2.5 x 10^24 FLOP. Training compute is the closest available proxy for how much was invested in a model, though it says nothing on its own about how well that compute was spent.
These two models share no benchmark on which both have been scored, so no direct performance comparison is possible here. The specification table below is a comparison of inputs, not of results.
| Field | Flan-PaLM 540B | GPT-4 (Jun 2023) |
|---|---|---|
| Organization | OpenAI | |
| Published | Oct 20, 2022 | Jun 13, 2023 |
| Training compute | 2.5 x 10^24 FLOP | 2.1 x 10^25 FLOP |
| Parameters | 540B | 1.8T |
| Dataset size | 1.4B | 5.4T |
| Training hardware | Google TPU v4 | NVIDIA A100 SXM4 40 GB |
| Chips used | 512 | 25,000 |
| Training time | 37 h | 2.3K h |
| Training power draw | 348.3 kW | 19.9 MW |
| Accessibility | Unreleased | API access |
| Open weights | No | No |
| Country | United States | United States |
