First try: flan-t5-base for representation model

app.py

@@ -6,17 +6,18 @@ import duckdb
 import numpy as np
 import requests
 
+from dotenv import load_dotenv
 from torch import cuda
 from gradio_huggingfacehub_search import HuggingfaceHubSearch
 from bertopic import BERTopic
 from bertopic.representation import KeyBERTInspired
+from bertopic.representation import TextGeneration
 from cuml.manifold import UMAP
 from cuml.cluster import HDBSCAN
 from huggingface_hub import HfApi
 from sklearn.feature_extraction.text import CountVectorizer
 from sentence_transformers import SentenceTransformer
-
-from dotenv import load_dotenv
+from transformers import pipeline
 
 # These imports at the end because of torch/datamapplot issue in Zero GPU
 # import spaces
@@ -51,10 +52,12 @@ CHUNK_SIZE = 10_000
 
 session = requests.Session()
 sentence_model = SentenceTransformer("all-MiniLM-L6-v2")
-keybert = KeyBERTInspired()
-vectorizer_model = CountVectorizer(stop_words="english")
 
-representation_model = KeyBERTInspired()
+prompt = "I have a topic described by the following keywords: [KEYWORDS]. Based on the previous keywords, what is this topic about?"
+generator = pipeline("text2text-generation", model="google/flan-t5-base")
+representation_model = TextGeneration(generator)
+
+vectorizer_model = CountVectorizer(stop_words="english")
 
 global_topic_model = None