تشغيل مسار ضمان الجودة

A Hugging Face pipeline wraps a model and tokenizer into one easy call. You hand it a question and context, it hands back an answer. 🚀

Everything starts with one import. The pipeline function from the transformers library builds the whole QA stack for you.

Pass the task name question-answering to pipeline. With no model named, it downloads a sensible default for you.

The pipeline needs a context: the passage that contains the answer. Think of it as the text the model is allowed to read.

Now call the pipeline with both a question and the context. It scans the passage and returns the best matching span.

The result is a dictionary. The answer text lives under the answer key, ready to print or store.

Each result also carries a score between 0 and 1. Higher means the model is more sure about its chosen span.

The result includes start and end character indexes. They tell you exactly where in the context the answer was found.

You can pin a specific model by name. A popular small QA model is distilbert fine-tuned on the SQuAD dataset.

One context can answer many questions. Build the pipeline once, then call it repeatedly with new questions on the same passage.

Remember: this extractive pipeline answers only from the context you give. Ask about something not in the text and you get a poor span.

Recall how the QA pipeline returns its result.

One pipeline call takes a question plus context and returns a dict with answer, score, start, and end. You ran extractive QA in three lines. ✅