Fail to resolve requirements.txt

[WhiteFlamesCN]

Hi there!
Thank you for your work! When I try to run "pip install -r requirements.txt" , it comes out a conflict between numpy and vllm.

BTW, I want to run this on a GPU machine(as Master node, for compute) and several CPU machines (as Slave node, for OSWorld ). Which steps are required to prepare the env on Master node , and the Slave node ?

Thanks for your time!

[FanbinLu]

requirements.txt fixed.

To prepare environment, you need to make sure:

1. All the machines share disk storage and you have to make sure you use the same conda environment for all nodes, otherwise the "ray start" might fail due to different ray versions.
2. Set passwordless ssh and configure docker group permissions(run without sudo). Ensure that each node can start OSWorld env with python-docker, see OSWorld/run_multienv_uitars.py
3. You might check the maximum envs you can create on a single node(in my case, 16 envs/node). Decrease the number if you find problems with the envs(docker collapse, communication error, a node gets offline, ...).