3 ms·
It sounds nice. How do you evaluate the performance of your way against usual embedding?
by higty 2y ago
It sounds nice.
How do you evaluate the performance of your way against usual embedding?
- Quizzical4230 2y agoBy assuming "usual embedding" meaning using the default model, which generally is "all-MiniLM-L6-v2", I used MixedBread's embedding model because of this [^1]. You can evaluate how well a model is doing by subjectively going through some search results for papers you have a good grasp on. Another way I look at is to see the 2D "maps" of the embeddings and how well these are segregated, see [^2]. [1]: https://www.mixedbread.ai/blog/binary-mrl https://www.mixedbread.ai/blog/binary-mrl [2]: https://raw.githubusercontent.com/mitanshu7/dumpyard/refs/heads/main/2024_kde_soft.png https://raw.githubusercontent.com/mitanshu7/dumpyard/refs/he...