QA-Pipeline: with or without Reader #7855
-
|
Hi, I'm new to haystack and NLP tasks and wondering: Why did Haystack 1.x recommend using a reader in the tutorial, but not Haystack 2.x (tutorial). Researching it my self ended poorly, because many articles are already outdated, and I can't decide what is right and wrong. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
In general, in the past (pre-ChatGPT era), Extractive Question Answering (using a Reader) was the standard. Nowadays, Generative Question Answering/RAG (using a Generator) is more popular and works better for most cases. For a thorough comparison between these 2 approaches, I recommend this article. Note that some of the links are to Haystack 1.x code, while we suggest to use 2.x. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you very much! That helped me. |
Beta Was this translation helpful? Give feedback.
In general, in the past (pre-ChatGPT era), Extractive Question Answering (using a Reader) was the standard.
Nowadays, Generative Question Answering/RAG (using a Generator) is more popular and works better for most cases.
For a thorough comparison between these 2 approaches, I recommend this article. Note that some of the links are to Haystack 1.x code, while we suggest to use 2.x.