Tumor segmentation from CT scans is a task typically performed by trained professionals in practice. However, the advent of vision language models (VLMs) present an alternative, machine-learning based approach to tackling this task. We explored the usage of MM-LLM-RO, a VLM, in the context of volume contouring for lung cancer as opposed to the initial author's intent of training the model on a dataset of breast cancer. We wanted to examine if the architecture overall could be extended to different types of tumors as well as if the same trained model weights could be utilized to segment multiple types of tumors. Although no state-of-the-art performances were matched, we discovered modest performances after training the model over a span of 1000 epochs, which took approximately 18 hours on 2 NVIDIA A40 GPUs. The training jobs on these NVIDIA GPUs were submitted largely with the SLURM job submission system on Penn Med’s CBICA cluster, though were initially submitted with the SGE method before Penn Med completed the migration.