Changes for the engine for using it in a pyodide context #2
Conversation
- moves data.py to cache.py ( I was having some conflicts with the old data folder) - Adding a check in the cache to see if pyodide_js is avaliable and if so changing the cache to use pyodide versions of the models - Adding a check to see if the models are bundled as Model or as a tuple and work appropriatly (I was having issues pickling the Model without using the demoland library. Bundling as a tuple of (W, model) seems more portable?) - Adding a build script for generating the wasm wheel TODO - Need to allow passing through of the BASE_URL for the pooch cache so that when this is hosted we can grab the files from the root URL
|
what is the state of this? some commented-out bits feel WIP? |
|
It needs a bit of a tidy but I wanted to get your eyes on it before we wrapped it up. Especially for the changes to model loading (tuple of weights and model vs the Model class) If it looks ok to you I can tidy up the code and we can merge it in |
|
Go for it. We can revise the model loading with the national model again as discussed before. |
|
Sounds good. Will try and do this today or tomorrow! |
|
@stuartlynn, are you ok if I merge this and reopen a new the PR from the wasm branch of this repo? I'd like to spend a little bit of time unifying all the different versions (fastAPI / Azure / WASM) into the same branch, otherwise it feels like the web app needs three different submodules which is pretty ugly. |
Adding some changes to make the engine work in wasm
TODO