How do I control the number of threads during the initialization of jaxlib/xla_extension.so？

[mochen302]

I am running my Python application in a Kubernetes cluster, and I am using the JAX library.

My container only requested 12 CPUs,

but I noticed that after successfully starting, hundreds of threads were initiated within my container.

Through GDB debugging, I found that most of these threads were launched by the following code.

I tried set the below envs, but it do not work!
XLA_FLAGS : --xla_force_host_platform_device_count=8 --xla_cpu_multi_thread_eigen=true intra_op_parallelism_threads=4 inter_op_parallelism_threads=4
Is it possible to manually control the number of initialization threads?