In BLS mode, does the data go to the CPU

[CallmeZhangChenchen]

model_name t2s_sdec platform: "tensorrt_plan"
I have a piece of BLS mode code here, I found that in the case of a relatively small number of cpu cores (120->24), the time will be doubled,
So I want to ask, is the data back to the CPU, not always on the GPU, is there any solution

[CallmeZhangChenchen]

I add parameters to t2s_sdec/config.json

parameters: { key: "FORCE_CPU_ONLY_INPUT_TENSORS" value: {string_value:"no"}}

There's still no change in speed

Suspect that BLS code inside the data ran to the CPU, but do not know how to optimize.
Because of previous experience, the number of cpu cores decreases, so the speed of cpu2gpu will slow down,

[CallmeZhangChenchen]

same as #7373 ! close