Update app.py

app.py

@@ -23,9 +23,9 @@ _ = load_dotenv(find_dotenv())
 
 
 # Schnittstellen hinzubinden und OpenAI Key holen aus den Secrets
-client = OpenAI(
-  api_key=os.getenv("OPENAI_API_KEY"),  # this is also the default, it can be omitted
-)
+#client = OpenAI(
+  #api_key=os.getenv("OPENAI_API_KEY"),  
+#)
 
 
 
@@ -41,8 +41,7 @@ client = OpenAI(
 #MONGODB_COLLECTION = client[MONGODB_DB_NAME][MONGODB_COLLECTION_NAME]
 #MONGODB_INDEX_NAME = "default"
 
-template = """If you don't know the answer, just say that you don't know, don't try to make up an answer. Keep the answer as concise as possible. Always say 
-              "🧠 Thanks for using the app - Bernd" at the end of the answer. """
+template = """If you don't know the answer, just say that you don't know, don't try to make up an answer. Keep the answer as concise as possible. Answer in german if not asked otherwise """
 
 llm_template = "Answer the question at the end. " + template + "Question: {question} Helpful Answer: "
 rag_template = "Use the following pieces of context to answer the question at the end. " + template + "{context} Question: {question} Helpful Answer: "
@@ -52,10 +51,11 @@ LLM_CHAIN_PROMPT = PromptTemplate(input_variables = ["question"],
 RAG_CHAIN_PROMPT = PromptTemplate(input_variables = ["context", "question"], 
                                   template = rag_template)
 
+OAI_API_KEY=os.getenv("OPENAI_API_KEY")
 #Pfad, wo Docs abgelegt werden können - lokal, also hier im HF Space (sonst auf eigenem Rechner)
 PATH_WORK = "."
-CHROMA_DIR  = "/data/chroma"
-YOUTUBE_DIR = "/data/youtube"
+CHROMA_DIR  = "/chroma"
+YOUTUBE_DIR = "/youtube"
 
 PDF_URL       = "https://arxiv.org/pdf/2303.08774.pdf"
 WEB_URL       = "https://openai.com/research/gpt-4"
@@ -75,10 +75,10 @@ def document_loading_splitting():
     loader = WebBaseLoader(WEB_URL)
     docs.extend(loader.load())
     # Load YouTube
-    #loader = GenericLoader(YoutubeAudioLoader([YOUTUBE_URL_1,
-                                               #YOUTUBE_URL_2,
-                                               #YOUTUBE_URL_3], YOUTUBE_DIR), 
-                           #OpenAIWhisperParser())
+    loader = GenericLoader(YoutubeAudioLoader([YOUTUBE_URL_1,
+                                               YOUTUBE_URL_2,
+                                               YOUTUBE_URL_3], PATH_WORK + YOUTUBE_DIR), 
+                           OpenAIWhisperParser())
     docs.extend(loader.load())
     # Document splitting
     text_splitter = RecursiveCharacterTextSplitter(chunk_overlap = 150,
@@ -102,8 +102,7 @@ def document_retrieval_chroma(llm, prompt):
     #Alternative Embedding - für Vektorstore, um Ähnlichkeitsvektoren zu erzeugen
     #embeddings = HuggingFaceInstructEmbeddings(model_name="sentence-transformers/all-MiniLM-L6-v2", model_kwargs={"device": "cpu"})
     db = Chroma(embedding_function = embeddings,
-                #persist_directory = CHROMA_DIR)
-                persist_directory = PATH_WORK + '/chroma')
+                persist_directory = PATH_WORK + CHROMA_DIR)
     
     return db
 
@@ -129,7 +128,8 @@ def rag_chain(llm, prompt, db):
 
 def invoke(openai_api_key, rag_option, prompt):
     if (openai_api_key == ""):
-        raise gr.Error("OpenAI API Key is required.")
+        #raise gr.Error("OpenAI API Key is required.")
+        openai_api_key= OAI_API_KEY
     if (rag_option is None):
         raise gr.Error("Retrieval Augmented Generation is required.")
     if (prompt == ""):
@@ -154,18 +154,17 @@ def invoke(openai_api_key, rag_option, prompt):
         raise gr.Error(e)
     return result
 
-description = """<strong>Overview:</strong> Reasoning application that demonstrates a <strong>Large Language Model (LLM)</strong> with 
-                 <strong>Retrieval Augmented Generation (RAG)</strong> on <strong>external data</strong>.\n\n
-                 <strong>Instructions:</strong> Enter an OpenAI API key and perform LLM use cases (semantic search, summarization, translation, etc.) on 
+description = """<strong>Überblick:</strong> Hier wird ein <strong>Large Language Model (LLM)</strong> mit 
+                 <strong>Retrieval Augmented Generation (RAG)</strong> auf <strong>externen Daten</strong> demonstriert.\n\n
+                 <strong>Genauer:</strong> Folgende externe Daten sind als Beispiel gegeben:
                  <a href='""" + YOUTUBE_URL_1 + """'>YouTube</a>, <a href='""" + PDF_URL + """'>PDF</a>, and <a href='""" + WEB_URL + """'>Web</a> 
-                 data on GPT-4, published after LLM knowledge cutoff.
+                 Alle neueren Datums!.
                  <ul style="list-style-type:square;">
-                 <li>Set "Retrieval Augmented Generation" to "<strong>Off</strong>" and submit prompt "What is GPT-4?" The <strong>LLM without RAG</strong> does not know the answer.</li>
-                 <li>Set "Retrieval Augmented Generation" to "<strong>Chroma</strong>" or "<strong>MongoDB</strong>" and submit prompt "What is GPT-4?" The <strong>LLM with RAG</strong> knows the answer.</li>
-                 <li>Experiment with prompts, e.g. "What are GPT-4's media capabilities in 5 emojis and 1 sentence?", "List GPT-4's exam scores and benchmark results.", or "Compare GPT-4 to GPT-3.5 in markdown table format."</li>
-                 <li>Experiment some more, for example "What is the GPT-4 API's cost and rate limit? Answer in English, Arabic, Chinese, Hindi, and Russian in JSON format." or "Write a Python program that calls the GPT-4 API."</li>
+                 <li>Setze "Retrieval Augmented Generation" auf  "<strong>Off</strong>" und gib einen Prompt ein." Das entspricht <strong> ein LLM nutzen ohne RAG</strong></li>
+                 <li>Setze "Retrieval Augmented Generation" to "<strong>Chroma</strong>"  und gib einen Prompt ein. Das <strong>LLM mit RAG</strong> weiß auch Antworten zu aktuellen Themen aus den angefügten Datenquellen</li>
+                 <li>Experimentiere mit Prompts, z.B.  Answer in German, Arabic, Chinese, Hindi, and Russian." oder "Schreibe ein Python Programm, dass die GPT-4 API aufruft."</li>
                  </ul>\n\n
-                 <strong>Technology:</strong> <a href='https://www.gradio.app/'>Gradio</a> UI using the <a href='https://openai.com/'>OpenAI</a> API and 
+                 <strong>Verwendete Technology:</strong> <a href='https://www.gradio.app/'>Gradio</a> UI using the <a href='https://openai.com/'>OpenAI</a> API and 
                  AI-native <a href='https://www.trychroma.com/'>Chroma</a> embedding database / 
                  <a href='https://www.mongodb.com/blog/post/introducing-atlas-vector-search-build-intelligent-applications-semantic-search-ai'>MongoDB</a> vector search. 
                  <strong>Speech-to-text</strong> (STT) via <a href='https://openai.com/research/whisper'>whisper-1</a> model, <strong>text embedding</strong> via