Posts

Post marked as solved
1 Replies
I finally found the source of this issue. Model is calling a function that has @torch.jit.script decorator: https://github.com/RVC-Project/Retrieval-based-Voice-Conversion-WebUI/blob/main/infer/lib/infer_pack/commons.py#L108 Removing this decorator, torch.jit.trace no longer add an intimplicit operator in the trace and CoreML tools could succesfully convert the model.
Post not yet marked as solved
3 Replies
Hi, I filed a Feedback for this old iOS frame capture issue. I understand that we should use latest version of macOS/Xcode/iOS for be able to debug/profile Metal code. However, our use case is to debug a problem that is only reproducible on iPhone 6s (for which max iOS available is 12.5.5). Disconnect/reboot/reconnect solves this issue for the iPhone 6S under 12.5.5. An other issue is MTLLibrary is crashing in newFunctionWithName or in internal readBytes (when committing a MTLCommandBuffer) when: metal library is compiled with source+driver info embedded (-MO or -frecord-sources option) frame capture is enabled iOS deployment version is lower than device iOS version Should I file a radar for this one ? Regards David