Questions about XLA Dump for Tensorflow models

[prrathi]

Hello,
I'm trying to capture HLO IR before and after optimization of some Tensorflow models by setting XLA_FLAGS = --xla_dump_to=./inputs/xla/{network}/ --xla_dump_hlo_as_text​. I'm noticing certain models emit several pairs of before_optimizations.txt​ and gpu_after_optimizations.txt​ dumps, with variable numbers across models, while others only have single. How can I determine how many sets will be dumped, and what does # sets dumped mean?
Also, I was curious about the numbers in the dumped files, what do the three numbers mean in these dumped file names?
module_0016.a_inference_run_step_1981__.1083.before_optimizations.txt module_0089.a_inference_run_step_1981__.1083.before_optimizations.txt​
Greatly appreciate any help that can be provided!