Flan-PaLM 540B (Google) and PaLM (540B) (Google Research) are both frontier AI models. Flan-PaLM 540B was published in October 2022 and PaLM (540B) in April 2022.
Flan-PaLM 540B was trained on 2.5 x 10^24 FLOP, essentially the same compute as 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 | PaLM (540B) |
|---|---|---|
| Organization | Google Research | |
| Published | Oct 20, 2022 | Apr 4, 2022 |
| Training compute | 2.5 x 10^24 FLOP | 2.5 x 10^24 FLOP |
| Parameters | 540B | 540.4B |
| Dataset size | 1.4B | 780B |
| Training hardware | Google TPU v4 | Google TPU v4 |
| Chips used | 512 | 6,144 |
| Training time | 37 h | 1.5K h |
| Training cost (2023 USD) | — | $3M |
| Training power draw | 348.3 kW | 4.2 MW |
| Accessibility | Unreleased | Unreleased |
| Open weights | No | No |
| Country | United States | United States |
