update readme

README.md

@@ -1327,15 +1327,17 @@ for doc in docs:
 Note, depending on the language, the download can be quite large.
 
 ## Search
-A full search example:
+A full search example (on the first 1,000 paragraphs):
 ```python
-#Run: pip install cohere datasets
+#Run: pip install cohere datasets numpy
 from datasets import load_dataset
-import torch
+import numpy as np
 import cohere
+import os
 
-co = cohere.Client(f"<<COHERE_API_KEY>>")  # Add your cohere API key from www.cohere.com
+co = cohere.Client("YOUR_COHERE_API_KEY")  # Add your cohere API key from www.cohere.com
 lang = "simple"
+top_k = 3
 
 #Load at max 1000 chunks + embeddings
 max_docs = 1000
@@ -1350,20 +1352,20 @@ for doc in docs_stream:
     if len(docs) >= max_docs:
         break
 
-doc_embeddings = torch.tensor(doc_embeddings)
+doc_embeddings = np.asarray(doc_embeddings)
 
-query = 'Who founded Youtube'
+query = 'Who was Alan Turing'
 response = co.embed(texts=[query], model='embed-multilingual-v3.0', input_type="search_query")
 query_embedding = response.embeddings 
-query_embedding = torch.tensor(query_embedding)
+query_embedding = np.asarray(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)
+dot_scores = np.matmul(query_embedding, doc_embeddings.transpose())[0]
+top_k_hits = np.argpartition(dot_scores, -top_k)[-top_k:]
 
 # Print results
 print("Query:", query)
-for doc_id in top_k.indices[0].tolist():
+for doc_id in top_k_hits:
     print(docs[doc_id]['title'])
     print(docs[doc_id]['text'])
     print(docs[doc_id]['url'], "\n")