Update funnel_search_with_matryoshka.ipynb

bootcamp/tutorials/milvus_usages/funnel_search_with_matryoshka.ipynb

@@ -816,9 +816,9 @@
    "metadata": {},
    "source": [
    "Here is a comparison of our search results across methods:\n",
-   "<div style='margin: auto; width: 50%;'><img src='results-raiders-of-the-lost-ark.png' width='100%'></div>\n",
-   "<div style='margin: auto; width: 75%;'><img src='results-ferris-buellers-day-off.png' width='100%'></div>\n",
-   "<div style='margin: auto; width: 50%;'><img src='results-the-shining.png' width='100%'></div>\n",
+   "<div style='margin: auto; width: 80%;'><img src='results-raiders-of-the-lost-ark.png' width='100%'></div>\n",
+   "<div style='margin: auto; width: 100%;'><img src='results-ferris-buellers-day-off.png' width='100%'></div>\n",
+   "<div style='margin: auto; width: 80%;'><img src='results-the-shining.png' width='100%'></div>\n",
     "We have shown how to use Matryoshka embeddings with Milvus for performing a more efficient semantic search algorithm called \"funnel search.\" We also explored the importance of the reranking and pruning steps of the algorithm, as well as a failure mode when the initial candidate list is too small. Finally, we discussed how the order of the dimensions is important when forming sub-embeddings - it must be in the same way for which the model was trained. Or rather, it is only because the model was trained in a certain way that prefixes of the embeddings are meaningful. Now you know how to implement Matryoshka embeddings and funnel search to reduce the storage costs of semantic search without sacrificing too much retrieval performance!"
    ]
   }