Update app.py

app.py

@@ -30,6 +30,8 @@ MONGODB_COLLECTION_NAME = "gpt-4"
 MONGODB_COLLECTION = client[MONGODB_DB_NAME][MONGODB_COLLECTION_NAME]
 MONGODB_INDEX_NAME = "default"
 
+description = os.environ["DESCRIPTION"]
+
 config = {
     "chunk_overlap": 150,
     "chunk_size": 1500,
@@ -183,30 +185,6 @@ def invoke(openai_api_key, rag_option, prompt):
         wandb_trace(rag_option, prompt, prompt_template, result, completion, chain_name, status_msg, start_time_ms, end_time_ms)
     return result
 
-description = """<strong>Overview:</strong> Context-aware multimodal reasoning application using a <strong>large language model (LLM)</strong> with 
-                 <strong>retrieval augmented generation (RAG)</strong>. 
-                 See the <a href='https://raw.githubusercontent.com/bstraehle/ai-ml-dl/main/hugging-face/openai-llm-rag.png'>architecture diagram</a>.\n\n
-                 <strong>Instructions:</strong> Enter an OpenAI API key and perform text generation use cases on <a href='""" + YOUTUBE_URL_1 + """'>YouTube</a>, 
-                 <a href='""" + PDF_URL + """'>PDF</a>, and <a href='""" + WEB_URL + """'>web</a> data published after LLM knowledge cutoff (example: GPT-4 data).
-                 <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 experiment with prompts. The <strong>LLM with RAG</strong> knows the answer:</li>
-                 <ol>
-                 <li>What are GPT-4's media capabilities in 5 emojis and 1 sentence?</li>
-                 <li>List GPT-4's exam scores and benchmark results.</li>
-                 <li>Compare GPT-4 to GPT-3.5 in markdown table format.</li>
-                 <li>Write a Python program that calls the GPT-4 API.</li>
-                 <li>What is the GPT-4 API's cost and rate limit? Answer in English, Arabic, Chinese, Hindi, and Russian in JSON format.</li>
-                 </ol>
-                 </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 
-                 AI-native <a href='https://www.trychroma.com/'>Chroma</a> embedding database or 
-                 <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> via <a href='https://openai.com/research/whisper'>whisper-1</a> model, <strong>text embedding</strong> via 
-                 <a href='https://openai.com/blog/new-and-improved-embedding-model'>text-embedding-ada-002</a> model, and <strong>text generation</strong> via 
-                 <a href='""" + WEB_URL + """'>gpt-4</a> model. Implementation via AI-first <a href='https://www.langchain.com/'>LangChain</a> toolkit. 
-                 Model performance evaluation via <a href='https://wandb.ai/bstraehle/openai-llm-rag/workspace'>Weights & Biases</a>."""
-
 gr.close_all()
 demo = gr.Interface(fn=invoke, 
                     inputs = [gr.Textbox(label = "OpenAI API Key", value = "sk-", lines = 1),