Update app.py

app.py

@@ -1 +1,24 @@
 import gradio as gr
+from faiss import IndexFlatIP
+import pandas as pd
+import numpy as np
+
+input_embeddings = np.load("bert_input_embeddings.npy")
+index = IndexFlatIP(input_embeddings.shape[-1])  
+index.add(input_embeddings)
+vocab = {v:k for k,v in tokenizer.vocab.items()}
+lookup_table = pd.Series(vocab).sort_index()
+
+def get_first_subword(word):
+    try:
+        return tokenizer.vocab[word]
+    except:
+        return tokenizer(word, add_special_tokens=False)['input_ids'][0]
+
+word_to_lookup = "##lity"
+num_neighbors = 20
+num_neighbors = 20
+i = get_first_subword(word_to_lookup)
+_ , I = index.search(input_embeddings[i:i+1], num_neighbors)
+hits = lookup_table.take(I[0])
+hits.values