FLAN 137B (Google Research) and Flan-PaLM 540B (Google) are both frontier AI models. FLAN 137B was published in September 2021 and Flan-PaLM 540B in October 2022.
Flan-PaLM 540B was trained on 2.5 x 10^24 FLOP, about 1.2x the compute of FLAN 137B at 2 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 137B | Flan-PaLM 540B |
|---|---|---|
| Organization | Google Research | |
| Published | Sep 3, 2021 | Oct 20, 2022 |
| Training compute | 2 x 10^24 FLOP | 2.5 x 10^24 FLOP |
| Parameters | 137B | 540B |
| Dataset size | 2.5T | 1.4B |
| Training hardware | Google TPU v3 | Google TPU v4 |
| Chips used | — | 512 |
| Training time | — | 37 h |
| Training power draw | — | 348.3 kW |
| Accessibility | Unreleased | Unreleased |
| Open weights | No | No |
| Country | United States | United States |
