Here’s the translation of the provided text into American English:
—
Amazon has launched two new user interface templates, called Text Ranking and Question and Answer, to enhance the experience for SageMaker AI customers. These tools are designed to optimize the quality of language models by enabling the collection of specific and structured user feedback.
The Text Ranking template allows human annotators to rank the responses generated by a large language model (LLM) according to custom criteria, such as relevance, clarity, or factual accuracy. This process is crucial for fine-tuning models using Reinforcement Learning from Human Feedback (RLHF), which improves the alignment of the model’s responses with user preferences. Meanwhile, the Question and Answer template facilitates the creation of high-quality question-and-answer pairs from provided text, serving as demonstration data for Supervised Fine-Tuning (SFT) and helping models respond more accurately to similar inputs.
To take advantage of these templates, users need to access the SageMaker AI console, which has introduced a new category for Generative AI. From this interface, it is possible to set up labeling jobs by specifying the location of the input manifest and the output path.
Using the Text Ranking template involves providing a detailed JSON file that describes the content to be ranked, allowing for structured evaluation. The annotated responses are stored in an S3 bucket chosen by the user, facilitating ongoing model assessment.
Regarding the Question and Answer template, it enables annotators to generate relevant questions and answers from text passages. This tool also offers a flexible format and a color-coded matching feature, helping annotators quickly identify the relevant sections of the text.
In addition to the graphical interface, users have access to a Labeling Job Creation API, which allows them to configure classification jobs programmatically, adding flexibility and integration into existing workflows.
With these innovations, Amazon SageMaker AI aims to empower its users to create high-quality datasets more efficiently, facilitating the training and evaluation of language models that better meet users’ needs and preferences.
—
Referrer: MiMub in Spanish
