Mert
Mert
This PR is a stripped down version of #2758 without Ray, batching or finer-grained configs. In particular, it modularizes the ML code and abstracts model processing into classes for each...
This PR integrates Ray Serve into the ML server. Motivation: - Optimized memory consumption. - Batching for faster inference. - More maintainable code structure. Features: - Easy process and thread...
Following the great work from #2563, I compiled the models to ONNX and cleaned up the Dockerfile and Python dependencies. I haven't done any serious profiling yet, but it seems...
### The bug I was looking through [validator](https://www.npmjs.com/package/validator) and noticed they have a `normalizeEmail` function that does more thorough sanitation beyond converting to lowercase. For instance, Gmail ignores dots so...
## Description This PR adds support for [Locust](https://locust.io/), a performance measurement tool. The benefit of this is being able to easily quantify and visualize ML performance. In turn, this makes...
## Description This PR adds `pytest` unit tests to the ML package. Given the expanding scope of ML, it's important to ensure a certain set of behaviors across changes. The...
## Description Aims to fix #2947 by adding error handling when loading models. If a model file is corrupt and fails to load, this will delete the folder for that...
## Description Switching to Express's `sendFile` results in much cleaner code that sets appropriate headers automatically. The current approach is too manual and error-prone in the absence of a good...
## Description WIP support for NVENC, Quick Sync and VAAPI transcoding backends. This will make transcoding much faster at the cost of lower quality at the same bitrate compared to...
### The bug The Store Videos Efficiently feature was introduced in Android 12 and is enabled by default on Google Pixel phones. In this setting, videos are stored as HEVC...