Update app.py

app.py

@@ -33,7 +33,7 @@ from PyPDF2 import PdfReader
 from templates import bot_template, css, user_template
 from xml.etree import ElementTree as ET
 
-# Llama Constants
+# 1. Constants and Top Level UI Variables
 API_URL = 'https://qe55p8afio98s0u3.us-east-1.aws.endpoints.huggingface.cloud'  # Dr Llama
 API_KEY = os.getenv('API_KEY')
 headers = {
@@ -44,11 +44,9 @@ key = os.getenv('OPENAI_API_KEY')
 prompt = f"Write instructions to teach anyone to write a discharge plan. List the entities, features and relationships to CCDA and FHIR objects in boldface."
 # page config and sidebar declares up front allow all other functions to see global class variables
 st.set_page_config(page_title="GPT Streamlit Document Reasoner", layout="wide")
-
-# UI Controls
 should_save = st.sidebar.checkbox("💾 Save", value=True, help="Save your session data.")
 
-# Function to add witty and humor buttons
+# 2. Prompt label button demo for LLM
 def add_witty_humor_buttons():
     with st.expander("Wit and Humor 🤣", expanded=True):
         # Tip about the Dromedary family
@@ -95,7 +93,7 @@ def add_witty_humor_buttons():
             StreamLLMChatResponse(descriptions["More Funny Rhymes 🎙️"])
 
 
-# Function to Stream Inference Client for Inference Endpoint Responses
+# 3. Stream Llama Response
 @st.cache_resource
 def StreamLLMChatResponse(prompt):
 
@@ -151,14 +149,8 @@ def StreamLLMChatResponse(prompt):
             <h1>🔊 Read It Aloud</h1>
             <textarea id="textArea" rows="10" cols="80">
         '''
-        
         documentHTML5 = documentHTML5 + result
-
         documentHTML5 = documentHTML5 + '''
-        This is a text passage that will be read aloud if Streamlit wrapping HTML5 wrapping browser based speech works correctly.
-        
-        If all works this html can be nested into a variable and then put into session using streamlit and HTML5.
-        
             </textarea>
             <br>
             <button onclick="readAloud()">🔊 Read Aloud</button>
@@ -166,31 +158,30 @@ def StreamLLMChatResponse(prompt):
         </html>
         '''
 
-
         import streamlit.components.v1 as components  # Import Streamlit
         components.html(documentHTML5, width=1280, height=1024)
-        
-        
         return result
     except:
-        st.write('DromeLlama is asleep. Starting up now on A10 - please give 5 minutes then retry as KEDA scales up from zero to activate running container(s).')
+        st.write('Llama model is asleep. Starting up now on A10 - please give 5 minutes then retry as KEDA scales up from zero to activate running container(s).')
 
+# 4. Run query with payload
 @st.cache_resource
 def query(payload):
     response = requests.post(API_URL, headers=headers, json=payload)
     st.markdown(response.json())
     return response.json()
-
 def get_output(prompt):
     return query({"inputs": prompt})
 
+# 5. Auto name generated output files from time and content
 def generate_filename(prompt, file_type):
     central = pytz.timezone('US/Central')
     safe_date_time = datetime.now(central).strftime("%m%d_%H%M")
     replaced_prompt = prompt.replace(" ", "_").replace("\n", "_")
-    safe_prompt = "".join(x for x in replaced_prompt if x.isalnum() or x == "_")[:90]
+    safe_prompt = "".join(x for x in replaced_prompt if x.isalnum() or x == "_")[:45]
     return f"{safe_date_time}_{safe_prompt}.{file_type}"
 
+# 6. Speech transcription via OpenAI service
 @st.cache_resource
 def transcribe_audio(openai_key, file_path, model):
     openai.api_key = openai_key
@@ -215,6 +206,7 @@ def transcribe_audio(openai_key, file_path, model):
         st.error("Error in API call.")
         return None
 
+# 7. Auto stop on silence audio control for recording WAV files
 def save_and_play_audio(audio_recorder):
     audio_bytes = audio_recorder(key='audio_recorder')
     if audio_bytes:
@@ -225,6 +217,7 @@ def save_and_play_audio(audio_recorder):
         return filename
     return None
 
+# 8. File creator that interprets type and creates output file for text, markdown and code
 @st.cache_resource
 def create_file(filename, prompt, response, should_save=True):
     if not should_save:
@@ -232,24 +225,26 @@ def create_file(filename, prompt, response, should_save=True):
     base_filename, ext = os.path.splitext(filename)
     has_python_code = bool(re.search(r"```python([\s\S]*?)```", response))
     if ext in ['.txt', '.htm', '.md']:
-        #with open(f"{base_filename}-Prompt.txt", 'w') as file:
-        #    file.write(prompt.strip())
-        #with open(f"{base_filename}-Response.md", 'w') as file:
-        #    file.write(response)
         with open(f"{base_filename}.md", 'w') as file:
-            file.write(response)
+            content = prompt.strip() + '\r\n' + response
+            file.write(content)
         
         if has_python_code:
             python_code = re.findall(r"```python([\s\S]*?)```", response)[0].strip()
+            # one copy with runnable section - tee up with python process shell
             with open(f"{base_filename}-Code.py", 'w') as file:
                 file.write(python_code)
+            # second copy with markdown of prompt and response
+            with open(f"{base_filename}.md", 'w') as file:
+                content = prompt.strip() + '\r\n' + response
+                file.write(content)
             
 def truncate_document(document, length):
     return document[:length]
-
 def divide_document(document, max_length):
     return [document[i:i+max_length] for i in range(0, len(document), max_length)]
 
+# 9. Sidebar with UI controls to review and re-run prompts and continue responses
 @st.cache_resource
 def get_table_download_link(file_path):
     with open(file_path, 'r') as file:
@@ -278,13 +273,15 @@ def get_table_download_link(file_path):
     href = f'<a href="data:{mime_type};base64,{b64}" target="_blank" download="{file_name}">{file_name}</a>'
     return href
 
+
 def CompressXML(xml_text):
     root = ET.fromstring(xml_text)
     for elem in list(root.iter()):
         if isinstance(elem.tag, str) and 'Comment' in elem.tag:
             elem.parent.remove(elem)
     return ET.tostring(root, encoding='unicode', method="xml")
-    
+
+# 10. Read in and provide UI for past files
 @st.cache_resource
 def read_file_content(file,max_length):
     if file.type == "application/json":
@@ -307,6 +304,7 @@ def read_file_content(file,max_length):
     else:
         return ""
 
+# 11. Chat with GPT - Caution on quota - now favoring fastest AI pipeline STT Whisper->LLM Llama->TTS
 @st.cache_resource
 def chat_with_model(prompt, document_section, model_choice='gpt-3.5-turbo'):
     model = model_choice
@@ -336,6 +334,7 @@ def chat_with_model(prompt, document_section, model_choice='gpt-3.5-turbo'):
     st.write(time.time() - start_time)
     return full_reply_content
 
+# 12. Embedding VectorDB for LLM query of documents to text to compress inputs and prompt together as Chat memory using Langchain
 @st.cache_resource
 def chat_with_file_contents(prompt, file_content, model_choice='gpt-3.5-turbo'):
     conversation = [{'role': 'system', 'content': 'You are a helpful assistant.'}]
@@ -368,6 +367,7 @@ def extract_file_extension(file):
     else:
         raise ValueError(f"Unable to extract file extension from {file_name}")
 
+# Normalize input as text from PDF and other formats
 @st.cache_resource
 def pdf2txt(docs):
     text = ""
@@ -390,11 +390,13 @@ def txt2chunks(text):
     text_splitter = CharacterTextSplitter(separator="\n", chunk_size=1000, chunk_overlap=200, length_function=len)
     return text_splitter.split_text(text)
 
+# Vector Store using FAISS
 @st.cache_resource
 def vector_store(text_chunks):
     embeddings = OpenAIEmbeddings(openai_api_key=key)
     return FAISS.from_texts(texts=text_chunks, embedding=embeddings)
 
+# Memory and Retrieval chains
 @st.cache_resource
 def get_chain(vectorstore):
     llm = ChatOpenAI()
@@ -428,6 +430,7 @@ def divide_prompt(prompt, max_length):
     chunks.append(' '.join(current_chunk))
     return chunks
 
+# 13. Provide way of saving all and deleting all to give way of reviewing output and saving locally before clearing it
 @st.cache_resource
 def create_zip_of_files(files):
     zip_name = "all_files.zip"
@@ -435,7 +438,6 @@ def create_zip_of_files(files):
         for file in files:
             zipf.write(file)
     return zip_name
-
 @st.cache_resource
 def get_zip_download_link(zip_file):
     with open(zip_file, 'rb') as f:
@@ -444,7 +446,7 @@ def get_zip_download_link(zip_file):
     href = f'<a href="data:application/zip;base64,{b64}" download="{zip_file}">Download All</a>'
     return href
 
-
+# 14. Inference Endpoints for Whisper (best fastest STT) on NVIDIA T4 and Llama (best fastest AGI LLM) on NVIDIA A10
 API_URL_IE = f'https://tonpixzfvq3791u9.us-east-1.aws.endpoints.huggingface.cloud'
 headers = {
 	"Authorization": "Bearer XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX",
@@ -465,7 +467,7 @@ def generate_filename(prompt, file_type):
     safe_prompt = "".join(x for x in replaced_prompt if x.isalnum() or x == "_")[:90]
     return f"{safe_date_time}_{safe_prompt}.{file_type}"
 
-# 10. Audio recorder to Wav file:
+# 15. Audio recorder to Wav file 
 def save_and_play_audio(audio_recorder):
     audio_bytes = audio_recorder()
     if audio_bytes:
@@ -475,7 +477,7 @@ def save_and_play_audio(audio_recorder):
         st.audio(audio_bytes, format="audio/wav")
         return filename
 
-# 9B. Speech transcription to file output - OPENAI Whisper
+# 16. Speech transcription to file output
 def transcribe_audio(filename):
     output = query(filename)
     return output
@@ -500,6 +502,8 @@ def whisper_main():
         create_file(filename, transcription, response, should_save)
         #st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True)
 
+
+# 17. Main
 def main():
 
     st.title("AI Drome Llama")
@@ -646,6 +650,7 @@ def main():
                 filename = generate_filename(raw, 'txt')
                 create_file(filename, raw, '', should_save)
 
+# 18. Run AI Pipeline
 if __name__ == "__main__":
     whisper_main()
     main()