Max Lund

If you want to load it anyway, there is a flag you can use as argument in automatic1111 webui launch.py: `--disable-safe-unpickle` Can be set in webui-user.sh (or .bat equivalent) e.g....

Any updates on this? Would love to run other CLIP variants via MLX, any advice would be appreciated!

I think this PR might shed some light on what is required to convert a SigLIP model to MLX, since @Blaizzy did it as part of adding PaliGemma to the...

> Yep the existing WhisperKit models are optimized for CoreML, the ones in[ this repo](https://huggingface.co/argmaxinc/whisperkit-mlx/tree/main) we will fill out with the equivalent weights that are compatible with this MLX PR...

@anishjain123 there is also this alternative, as you probably know already: [mlx-whisper](https://github.com/ml-explore/mlx-examples/tree/main/whisper) @ZachNagengast what would be the differences between running Whisper via MLX using your implementation vs the mlx-whisper version...

We haven't been able to pinpoint which file actually caused the issue, as we ran through thousands of files and weren't able to log an exception due to the crash....

However, we know that we can't reliably reproduce it since it occurred for the same set of MXF files that I've now successfully parsed using the MediaInfo lib.

I just managed to reproduce it by running two threads through a couple of thousand MXF files. These are the last lines of the application logs: ``` 2025-06-04 18:36:34,042 -...

Ok I can reliably reproduce it by simulating a bunch of concurrent calls: ``` import os from concurrent.futures import ThreadPoolExecutor, as_completed from pymediainfo import MediaInfo MEDIAINFO_LIB_PATH = os.path.expanduser('~/libmediainfo.0.dylib') N_THREADS =...

Aha, thanks @g-maxime! So the .MXF files were a red herring. Sounds like I can just add the two lines calling `MediaInfo_Option` once during application startup, and remove the concurrency...