Merve developed and integrated support for using the Hugging Face Transformers backend with vLLM online serving in the huggingface/trl repository. This work involved backend development and Python scripting to enable users to select the vLLM model implementation and seamlessly switch between backends. Merve updated documentation and setup scripts, improving onboarding and operational clarity for users deploying models online. The integration allows for greater flexibility and scalability in model serving, addressing the need for easier backend selection. The depth of the work is reflected in the cohesive updates to both code and documentation, ensuring a smoother experience for both developers and users.