Do optimum-benchmark support deepspeed_mii

[neozhang307]

Thank you very much for your efforts in this work.

I found that you support deepspeed. But I think for inference workload, deepspeed-mii has better performance, as claimed in this repo: https://github.com/microsoft/DeepSpeed-MII.

I am curious about its performance. I wonder how hard it is to support this backend?

[IlyasMoutawwakil]

nope we don't but you can submit a PR following the example of in #250

[neozhang307]

Let me see if I can implement it myself. But I don't know where to start.
The basic API for mii is:

import mii
pipe = mii.pipeline("mistralai/Mistral-7B-v0.1")
response = pipe(["DeepSpeed is", "Seattle is"], max_new_tokens=128)
print(response)

Can you point out some backend code that I can start from?