llava-hf/llava-1.5-7b-hf · The environment settings for fine-tuning the llava-1.5-7b-hf

I am trying to fine-tune the llava-1.5-7b-hf model using the sft_vlm.py script from the TRL library. However, I keep encountering compatibility issues with DeepSpeed, such as version mismatches and environment errors. Could someone provide guidance on the recommended environment setup for this task? Specifically, I would appreciate details on the versions of PyTorch, CUDA, DeepSpeed, and any other dependencies that are required to run this script smoothly. Any tips or best practices for avoiding such issues would also be helpful. Thank you!