Log_likelihood can be pretty big. The memory footprint is chains * draws * datasize * data dtype.
Some discussion here: compute_log_likelihood for large datasets · Issue #6864 · pymc-devs/pymc · GitHub
1 Like