Dockerfile now loads models to local folder. Can use custom output folder. requrirements for GPU-enabled summarisation now in separate file to hopefully avoid HF space issues.

Upgraded Gradio version. Changed large model to Phi 3 128k. Now uses Llama cpp Python