Open daavoo opened 1 year ago
So looks like the plan could be to:
async
version of async_post_live_metrics
._post_in_chunks
async. Send params
and metrics
async, process data and send the image / plots chunk async as well. post_live_metrics
, accept a blocking: Bool = True
. By sticking with the current behavior by default, we can keep the Bool
return type for existing client code. In non-blocking mode, we can return asyncio.create_task(async_post_live_metrics(...))
.Am I close? If that works, happy to submit a PR with that, unit tests, and docs!
Now that images can be sent through the API, the requests can take significant time, and would be nice to not block the main thread where the training will be happening.