Skip to content

Commit

Permalink
Merge branch 'master' into thomasht86/fix-vespacli-workflow
Browse files Browse the repository at this point in the history
  • Loading branch information
thomasht86 authored Nov 18, 2024
2 parents 5aef1ee + 6ab0238 commit 4718a95
Show file tree
Hide file tree
Showing 4 changed files with 5 additions and 3 deletions.
Binary file added docs/sphinx/source/_static/sample_1.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Binary file added docs/sphinx/source/_static/sample_2.png
Loading
Sorry, something went wrong. Reload?
Sorry, we cannot display this file.
Sorry, this file is invalid so it cannot be displayed.
Original file line number Diff line number Diff line change
Expand Up @@ -15,10 +15,12 @@
"\n",
"# Exploring the potential of OpenAI Matryoshka 🪆 embeddings with Vespa\n",
"\n",
"This notebook demonstrates the effectiveness of using the recently released(As of January 2024) OpenAI `text-embedding-3` embeddings with Vespa.\n",
"This notebook demonstrates the effectiveness of using the recently released (as of January 2024) OpenAI `text-embedding-3` embeddings with Vespa.\n",
"\n",
"Specifically, we are interested in the [Matryoshka Representation Learning](https://aniketrege.github.io/blog/2024/mrl/) technique used in training, which lets us \"shorten embeddings (i.e. remove some numbers from the end of the sequence) without the embedding losing its concept-representing properties\". This allow us to trade off a small amount of accuracy in exchange for much smaller embedding sizes, so we can store more documents and search them faster.\n",
"\n",
"[Exploring the potential of OpenAI Matryoshka 🪆 embeddings with Vespa](https://blog.vespa.ai/matryoshka-embeddings-in-vespa/) and [Matryoshka 🤝 Binary vectors: Slash vector search costs with Vespa](https://blog.vespa.ai/combining-matryoshka-with-binary-quantization-using-embedder/) are good reads on this subject.\n",
"\n",
"By using [phased ranking](https://docs.vespa.ai/en/phased-ranking.html), we can re-rank the top K results with the full embeddings in a second step. This produces accuracy on par with using the full embeddings!\n",
"\n",
"We'll use a standard information retrieval benchmark to evaluate result quality with different embedding sizes and retrieval/ranking strategies.\n",
Expand Down
Original file line number Diff line number Diff line change
Expand Up @@ -456,8 +456,8 @@
"These reports are the ones we are going to use for this showcase.\n",
"Here are some sample images:\n",
"\n",
"![Sample1](./static/assets/page_95.png)\n",
"![Sample2](./static/assets/page_74.png)\n"
"![Sample1](../_static/sample_1.png)\n",
"![Sample2](../_static/sample_2.png)\n"
]
},
{
Expand Down

0 comments on commit 4718a95

Please sign in to comment.