Update README.md

README.md

@@ -1,4 +1,47 @@
 # myanimelist-embeddings
 
 This dataset is every non-empty anime synopsis from [MyAnimeList.net](https://myanimelist.net) ran
-through the `embed-multilingual-v2.0` embedding model from [Cohere AI](https://cohere.com).
+through the `embed-multilingual-v2.0` embedding model from [Cohere AI](https://cohere.com).
+
+## Sample code for searching the data set
+
+```python
+import os
+
+import cohere
+import torch
+from datasets import load_dataset
+
+co = cohere.Client(
+    os.environ["COHERE_API_KEY"]
+)  # Add your cohere API key from www.cohere.com
+
+docs_stream = load_dataset(
+    f"abatilo/myanimelist-embeddings", split="train", streaming=True
+)
+
+docs = []
+doc_embeddings = []
+
+for doc in docs_stream:
+    docs.append(doc)
+    doc_embeddings.append(doc["embedding"])
+
+doc_embeddings = torch.tensor(doc_embeddings)
+
+while True:
+    query = input("Enter query: ")
+    print("Query:", query)
+
+    response = co.embed(texts=[query], model="embed-multilingual-v2.0")
+    query_embedding = response.embeddings
+    query_embedding = torch.tensor(query_embedding)
+
+    # Compute dot score between query embedding and document embeddings
+    dot_scores = torch.mm(query_embedding, doc_embeddings.transpose(0, 1))
+    top_k = torch.topk(dot_scores, k=3)
+
+    for doc_id in top_k.indices[0].tolist():
+        print(docs[doc_id]["title"])
+        print(docs[doc_id]["synopsis"], "\n")
+```