Multi-instance TPM/RPM control without response caching

[hula-la]

I would like to inquire about the possibility of key-based TPM & RPM control in a multi-instance environment.
After examining the code, I found that since it uses local cache by default, TPM & RPM are not shared across multiple instances, so control doesn't work as expected.
I confirmed that setting default_in_memory_ttl to 0 forces the use of TPM & RPM stored in Redis, which results in normal control as expected.
However, when setting cache=true, there's an issue where responses are also cached unintentionally.
I want to control only TPM & RPM without caching responses. Is this configuration possible?

Thank you:)

[hula-la]

EXPERIMENTAL_MULTI_INSTANCE_RATE_LIMITING is supported
https://docs.litellm.ai/docs/proxy/users#beta-multi-instance-rate-limiting