You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I use my own labeled dataset and I want to train SLEAP with multi-animal bottom pipeline. I go through the GUI to train.
Expected behaviour
Training works.
Actual behaviour
I have an error at the very beginning of the training
[...]
INFO:sleap.nn.training:Created run path: /Users/dcclab/Documents/vpineaunoel/GitHub/AutomaticBehaviourAnalysis/mice_vivarium_sleap/models/20241104-MultianimalBottomUp241104_110233.multi_instance.n=1540
INFO:sleap.nn.training:Setting up visualization...
python(36403,0x7ff855b867c0) malloc: *** error for object 0x600005d481a4: pointer being freed was not allocated
python(36403,0x7ff855b867c0) malloc: *** set a breakpoint in malloc_error_break to debug
Run Path: /Users/dcclab/Documents/vpineaunoel/GitHub/AutomaticBehaviourAnalysis/mice_vivarium_sleap/models/20241104-MultianimalBottomUp241104_110233.multi_instance.n=1540
I thought it was a problem with the memory so I changed the maximal number of threads to 16 before launching the SLEAP GUI with export OMP_NUM_THREADS=16 but I still have the error.
Your personal set up
OS: MacOS Pro 2019
Processor : 3.2 GHz 16-Core Intel Xeon W
Memory : 768 GB 2933 MHz DDR4
GPU : AMD Radeon Pro Vega II, 32 GB of VRAM
Version(s): SLEAP 1.3.3, python 3.9.15, tensorflow 2.12.0 (I actually installed tensorflow-metal to use my GPU), numpy 1.23.00 (I changed that to make all dependencies work)
We actually don't support training on anything besides NVIDIA GPUs or Silicon Macs.
We can make this more explicit in the documentation in the future.
While SLEAP is installable on Pre-M1 (MacOS without Silicon GPUs) we do not expect training to work as is.
Our recommendation is to have a workstation with a NVIDIA GPU for training models. Depending on how difficult your models are to train, a workstation with around 24 GB of GPU memory is a real asset for a project. If you are not training frequently or if your models aren't too difficult to train, 8 GB of GPU memory could be sufficient.
The Apple Mini's with the latest Silicon GPU is another good option. With the Silicon GPU hardware changing rapidly our team will have to prioritize the latest hardware to maintain compatibility with our software.
Bug description
I use my own labeled dataset and I want to train SLEAP with multi-animal bottom pipeline. I go through the GUI to train.
Expected behaviour
Training works.
Actual behaviour
I have an error at the very beginning of the training
I thought it was a problem with the memory so I changed the maximal number of threads to 16 before launching the SLEAP GUI with
export OMP_NUM_THREADS=16
but I still have the error.Your personal set up
Environment packages
mamba list :
Logs
Screenshots
How to reproduce
sleap-label
The text was updated successfully, but these errors were encountered: