loc(“mps_select”(“(mpsFileLoc): /AppleInternal/Library/BuildRoots/75428952-3aa4-11ee-8b65-46d450270006/Library/Caches/com.apple.xbs/Sources/MetalPerformanceShadersGraph/mpsgraph/MetalPerformanceShadersGraph/Core/Files/MPSGraphUtilities.mm”:294:0)): error: ‘anec.gain_offset_control’ op result #0 must be 4D/5D memref of 16-bit float or 8-bit signed integer or 8-bit unsigned integer values, but got ‘memref<1x1x1x1xi1>’
Haa anyone else encounter this? it only started happening after the OS upgrade
I had to uninstall my entire tensorflow, conda, dependecies and reinstall from a clean slate - that solved the problem but opened a new one. I will post it separattely
Do you happen to resolve the issue? I had been training with a older version of tensorflow before upgrading to Sonoma on the M1 max. Now I have the same issue with training. Can you please share your setup and some recommendations? Thanks!
I use tf.config.set_visible_devices(, ‘GPU’)
this pretty much forces the code to use CPU and no GPUs – seems to get me the speed I needed.
I saw forms claim that the GPU is only designed for heavy load with large batch_size of 1,200, or 2,400 I use 64 for my batch size and the CPU gives me plenty of speed
I actually just left it alone after the error showed up (set to use GPU), and it eventually started training, so perhaps it is a non-issue. I think trying with smaller batch size will slow my training, which is already slow with GPU…