I stitched together multiple currency records from historical data to produce a fairly comprehensive dataset. I cut out some currencies that no longer exist and were of limited value and those that there weren't consistent records for.
Time series forecasting of currencies is notoriously difficult and this would be a good dataset to practice on. Ordinarily it is difficult to beat a random walk with forecasting. Maybe you can do better? Pick a currency and try.
Source: Reserve Bank of Australia 2022