Hello everyone! I have come across some unexpectedly high memory usage whilst running a hierarchical model - so much so my AWS Linux container (12GB memory) is running out of memory. Set-up: 2.1m observations Params by tree level Level 0 - 3 params Level 1 - 151 params Level 2 - 576 params L…

@cocodimama I think the bottleneck is the logp/dlogp function which holds several intermediate computations in RAM. It’s hard to say what freedom you have, because this depends on the exact model and operations you have. There is a memory profiler for PyTensor functions but I have never used it mys…

Understand root cause of high memory utilization

Questions v3

ricardoV94 December 6, 2023, 9:48pm 2

Yes I think it’s the number of parameters. I can imagine the computational graph (logp and gradient) easily taking what’s left of your RAM.

Topic		Replies	Views
Memory Error when sampling my Hierarchical logistic model Questions	8	1592	March 17, 2021
Memory spike at the end of the MCMC sampling Questions	10	2681	July 15, 2022
Compute_log_likelihood RAM usage is huge version agnostic arviz	4	524	September 27, 2023
Excessive memory - Multiple regression Questions	9	2931	October 24, 2017
GPU utilization is high but memory usage is very low leading to subpar sampling performance Questions gpu , theano	2	6233	August 26, 2017

Understand root cause of high memory utilization

Related topics