Add overlap and reduce search count.

app.py

@@ -9,10 +9,13 @@ from langchain.vectorstores import FAISS
 
 
 # Number of search results to query from the vector database.
-SIMILARITY_SEARCH_COUNT = 7
+SIMILARITY_SEARCH_COUNT = 3
 
 # Size of each document chunk in number of characters.
-CHUNK_SIZE = 500
+CHUNK_SIZE = 800
+
+# Chunk overlap in number of characters.
+CHUNK_OVERLAP = 50
 
 # Maximum number of output tokens.
 MODEL_MAX_LENGTH = 500
@@ -23,7 +26,7 @@ loader = PyMuPDFLoader("rdna3-shader-instruction-set-architecture-feb-2023_0.pdf
 documents = loader.load()
 
 print("Creating chunks")
-splitter = RecursiveCharacterTextSplitter(chunk_size=CHUNK_SIZE, chunk_overlap=0)
+splitter = RecursiveCharacterTextSplitter(chunk_size=CHUNK_SIZE, chunk_overlap=CHUNK_OVERLAP)
 chunks = splitter.split_documents(documents)
 
 print("Creating database")