This PR refines the existing detection for chat input in Text Generation Pipeline. Now the pipeline can correctly detect if the input is a chat input, and apply chat templates to it. It also modifies the pipeline to optionally accept a parameter that controls if it returns the full text as input or not.
It also includes a refactor to the Streamer class, making it more streamlined, and responsive to the output of text generation models.
What:
Description:
This PR refines the existing detection for chat input in Text Generation Pipeline. Now the pipeline can correctly detect if the input is a chat input, and apply chat templates to it. It also modifies the pipeline to optionally accept a parameter that controls if it returns the full text as input or not.
It also includes a refactor to the Streamer class, making it more streamlined, and responsive to the output of text generation models.