Update dataset with title/texts & 1m -> 41m in markdown

app.py

@@ -9,7 +9,7 @@ import faiss
 from usearch.index import Index
 
 # Load titles and texts
-title_text_dataset = load_dataset("mixedbread-ai/wikipedia-embed-en-2023-11", split="train").select_columns(["title", "text"])
+title_text_dataset = load_dataset("mixedbread-ai/wikipedia-data-en-2023-11", split="train", num_proc=4).select_columns(["title", "text"])
 
 # Load the int8 and binary indices. Int8 is loaded as a view to save memory, as we never actually perform search with it.
 int8_view = Index.restore("wikipedia_int8_usearch_50m.index", view=True)
@@ -75,15 +75,15 @@ with gr.Blocks(title="Quantized Retrieval") as demo:
     gr.Markdown(
 """
 ## Quantized Retrieval - Binary Search with Scalar (int8) Rescoring
-This demo showcases the retrieval using [quantized embeddings](https://huggingface.co/blog/embedding-quantization). The corpus consists of 1 million texts from Wikipedia articles.
+This demo showcases exact retrieval using [quantized embeddings](https://huggingface.co/blog/embedding-quantization). The corpus consists of 41 million texts from Wikipedia articles.
 
 <details><summary>Click to learn about the retrieval process</summary>
 
 Details:
 1. The query is embedded using the [`mixedbread-ai/mxbai-embed-large-v1`](https://huggingface.co/mixedbread-ai/mxbai-embed-large-v1) SentenceTransformer model.
 2. The query is quantized to binary using the `quantize_embeddings` function from the SentenceTransformers library.
-3. A binary index (1M binary embeddings; 130MB of memory/disk space) is searched using the quantized query for the top 40 documents.
-4. The top 40 documents are loaded on the fly from an int8 index on disk (1M int8 embeddings; 0 bytes of memory, 1.19GB of disk space).
+3. A binary index (41M binary embeddings; 5.2GB of memory/disk space) is searched using the quantized query for the top 40 documents.
+4. The top 40 documents are loaded on the fly from an int8 index on disk (41M int8 embeddings; 0 bytes of memory, 47.5GB of disk space).
 5. The top 40 documents are rescored using the float32 query and the int8 embeddings to get the top 10 documents.
 6. The top 10 documents are sorted by score and displayed.