update app.py

app.py

@@ -7,13 +7,15 @@ from openai import OpenAI
 import random
 import prompts
 
+# Set the cache directory to persistent storage
+os.environ["HF_HOME"] = "/data/.cache/huggingface"
 
 # client = OpenAI(api_key=st.secrets["general"]["OpenAI_API"])
 client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
 
-st.set_page_config(page_title="The AI Teaching Assistant")
 
-def generate_YT_file_names(base_path, embedding_model_name, chunk_tokens, overlap_tokens):
+@st.cache_resource
+def load_youtube_data(base_path, embedding_model_name, chunk_tokens, overlap_tokens):
     embedding_space_file_name = f'{base_path}/yt_embedding_space_{embedding_model_name}_tpc{chunk_tokens}_o{overlap_tokens}.json'
     with open(embedding_space_file_name, 'r') as json_file:
         loaded_data = json.load(json_file)
@@ -21,7 +23,8 @@ def generate_YT_file_names(base_path, embedding_model_name, chunk_tokens, overla
     embedding_space = np.array(loaded_data['embedding_space'])
     return loaded_data['chunks'], embedding_space
 
-def generate_Latex_file_names(base_path, embedding_model_name, chunk_tokens, overlap_tokens):
+@st.cache_resource
+def load_book_data(base_path, embedding_model_name, chunk_tokens, overlap_tokens):
     embedding_space_file_name = f'{base_path}/latex_embedding_space_by_sections_{embedding_model_name}_tpc{chunk_tokens}_o{overlap_tokens}.json'
     with open(embedding_space_file_name, 'r') as json_file:
         loaded_data = json.load(json_file)
@@ -29,21 +32,6 @@ def generate_Latex_file_names(base_path, embedding_model_name, chunk_tokens, ove
     embedding_space = np.array(loaded_data['embedding_space'])
     return loaded_data['chunks'], embedding_space
 
-@st.cache_resource
-def load_text_data(json_file_name):
-    with open(json_file_name, 'r') as f:
-        data = json.load(f)
-    return data
-
-
-@st.cache_resource
-def load_embeddings(npy_file_name):
-    return np.load(npy_file_name)
-
-@st.cache_resource
-def load_model(model_name):
-    return SentenceTransformer(model_name)
-
 @st.cache_resource
 def load_summary(file_path):
     with open(file_path, 'r') as file:
@@ -118,16 +106,15 @@ disclaimer_contact =""":gray[AI Teaching Assistant is developed at the Universit
 """
 
 # ---------------------------------------
-
-
 base_path = "data/"
 
-st.title(":red[AI Teaching Assistant]")
-# st.markdown("### Finite Element Methods")
+st.set_page_config(page_title="AI University")
+st.title(":red[AI University]")
+st.markdown("### Finite Element Methods")
 # st.markdown("### Based on Introduction to Finite Element Methods (FEM) by Prof. Krishna Garikipati")
 # st.markdown("##### [YouTube playlist of the FEM lectures](https://www.youtube.com/playlist?list=PLJhG_d-Sp_JHKVRhfTgDqbic_4MHpltXZ)")
 
-st.markdown(":gray[Welcome to] :red[AI Teaching Assistant]:gray[, developed at the] :red[University of Southern California]:gray[. This app leverages AI to provide expert answers to queries related to] :red[Finite Element Methods (FEM)]:gray[.]")
+st.markdown(":gray[Welcome to] :red[AI University]:gray[, developed at the] :red[University of Southern California]:gray[. This app leverages AI to provide expert answers to queries related to] :red[Finite Element Methods (FEM)]:gray[.]")
 
 # As the content is AI-generated, we strongly recommend independently verifying the information provided. 
 
@@ -202,7 +189,6 @@ with st.sidebar:
         integration_top_p = st.slider("Top P", 0.1, 0.5, .3, help="Defines the range of token choices the model can consider in the next prediction. Lower: More focused and restricted to high-probability options. Higher: More creative, allowing consideration of less likely options.", key='a2p')
 
         
-
 # Main content area
 if "question" not in st.session_state:
     st.session_state.question = ""
@@ -240,8 +226,8 @@ with col2:
         )
 
 # Load YouTube and LaTeX data
-text_data_YT, context_embeddings_YT = generate_YT_file_names(base_path, model_name, yt_chunk_tokens, yt_overlap_tokens)
-text_data_Latex, context_embeddings_Latex = generate_Latex_file_names(base_path, model_name, latex_chunk_tokens, latex_overlap_tokens)
+text_data_YT, context_embeddings_YT = load_youtube_data(base_path, model_name, yt_chunk_tokens, yt_overlap_tokens)
+text_data_Latex, context_embeddings_Latex = load_book_data(base_path, model_name, latex_chunk_tokens, latex_overlap_tokens)
 
 summary = load_summary('data/KG_FEM_summary.json')