Traditional RAG vs. HyDE, clearly explained (with visuals):
Questions are not semantically similar to their answers.
As a result, several irrelevant contexts get retrieved due to a higher cosine similarity.
As a result, several irrelevant contexts get retrieved due to a higher cosine similarity.
HyDE attempts to solve this.
The following visual depicts how it differs from traditional RAG and HyDE.
The following visual depicts how it differs from traditional RAG and HyDE.
- Use an LLM to generate a hypothetical answer H for the query Q.
- Embed the answer using a contriever model to get E.
- Query the vector database and fetch relevant context (C) using E.
- Pass the retrieved-context C + query Q to the LLM to produce an answer.
- Embed the answer using a contriever model to get E.
- Query the vector database and fetch relevant context (C) using E.
- Pass the retrieved-context C + query Q to the LLM to produce an answer.
of course, the hypothetical generated will contain hallucinated details.
The contriever model tries to handle this.
It is trained using contrastive learning to function as a near-lossless compressor whose task is to filter out the hallucinated details.
The contriever model tries to handle this.
It is trained using contrastive learning to function as a near-lossless compressor whose task is to filter out the hallucinated details.
Several studies have shown that HyDE improves the retrieval performance compared to the traditional embedding model.
But this comes at the cost of increased latency and more LLM usage.
But this comes at the cost of increased latency and more LLM usage.
That's a wrap!
If you enjoyed this tutorial:
Find me → @_avichawla
Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.
If you enjoyed this tutorial:
Find me → @_avichawla
Every day, I share tutorials and insights on DS, ML, LLMs, and RAGs.
• • •
Missing some Tweet in this thread? You can try to
force a refresh
