CUDA device rank in mii.pipeline

[RealPolitiX]

Is there a way to specify cuda device rank in mii.pipeline? I'm currently using deepspeed-mii version 0.2.3, which is released in March 2024. I tried adding model_config={"device_map":[1]}, but this still runs on the first gpu (rank 0).

The full line is mii.pipeline("mistralai/Mistral-7B-Instruct-v0.1", model_config={"device_map":[1]})