Although the .whl files are cached between code runs, the fact that we need to reset pyodide between runs means that they need to be re-interpreted each time the code runs. This introduces some latency which we may be able to reduce if we cache using WebAssembly modules (see Slack thread for more details)
There is a noticeable delay of around 5 seconds at the start of each code run using sklearn as a (rather large) example, but this is just over a second for something like numpy (which perhaps we are more likely to use)
It may be possible to cache the wasm files, but this thread suggests it's complicated and may not make enough to difference to be worth it
Although the
.whl
files are cached between code runs, the fact that we need to resetpyodide
between runs means that they need to be re-interpreted each time the code runs. This introduces some latency which we may be able to reduce if we cache using WebAssembly modules (see Slack thread for more details)