GMC: response timeout with some specific models

[KfreeZ]

In GMC e2e tests, there are some failed case caused by response timeout,

tgi2.2.0 + meta-llama/CodeLlama-7b-hf on xeon in codegen test

tgi2.2.0 + meta-llama/CodeLlama-7b-hf on guadi is ok

switch to another model "HuggingFaceH4/mistral-7b-grok" also ok

due to the manifest test (without GMC) cannot detect this issue, we need to find out whether it's GMC's specific.

[lianhao]

on gaudi, we use tgi-gaudi 2.0.1, not tgi2.2.0

[KfreeZ]

similar issue to #258, close this one as duplication