Spaces:

awacke1
/

LlamaWhisperer

Runtime error

App Files Files Community

awacke1 commited on Sep 24, 2023

Commit

7af9d8f

1 Parent(s): 1b6dd3a

Update app.py

Browse files

Files changed (1) hide show

app.py +189 -55

app.py CHANGED Viewed

@@ -33,22 +33,79 @@ from PyPDF2 import PdfReader
 from templates import bot_template, css, user_template
 from xml.etree import ElementTree as ET
-# Llama Constants
-API_URL = 'https://qe55p8afio98s0u3.us-east-1.aws.endpoints.huggingface.cloud'  # Dr Llama
 API_KEY = os.getenv('API_KEY')
 headers = {
-    "Authorization": f"Bearer {API_KEY}",
     "Content-Type": "application/json"
 }
 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
 def add_witty_humor_buttons():
     with st.expander("Wit and Humor 🤣", expanded=True):
         # Tip about the Dromedary family
@@ -94,8 +151,40 @@ def add_witty_humor_buttons():
         if col7[0].button("More Funny Rhymes 🎙️"):
             StreamLLMChatResponse(descriptions["More Funny Rhymes 🎙️"])
-# Function to Stream Inference Client for Inference Endpoint Responses
 def StreamLLMChatResponse(prompt):
     try:
@@ -132,27 +221,27 @@ def StreamLLMChatResponse(prompt):
             except:
                 st.write('Stream llm issue')
-        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).')
 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})
 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]
     return f"{safe_date_time}_{safe_prompt}.{file_type}"
 def transcribe_audio(openai_key, file_path, model):
     openai.api_key = openai_key
     OPENAI_API_URL = "https://api.openai.com/v1/audio/transcriptions"
@@ -176,6 +265,7 @@ def transcribe_audio(openai_key, file_path, model):
         st.error("Error in API call.")
         return None
 def save_and_play_audio(audio_recorder):
     audio_bytes = audio_recorder(key='audio_recorder')
     if audio_bytes:
@@ -186,34 +276,35 @@ def save_and_play_audio(audio_recorder):
         return filename
     return None
 def create_file(filename, prompt, response, should_save=True):
     if not should_save:
         return
     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)
         if has_python_code:
             python_code = re.findall(r"```python([\s\S]*?)```", response)[0].strip()
             with open(f"{base_filename}-Code.py", 'w') as file:
                 file.write(python_code)
 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)]
 def get_table_download_link(file_path):
     with open(file_path, 'r') as file:
-        try:
-            data = file.read()
-        except:
-            st.write('')
-            return file_path
     b64 = base64.b64encode(data.encode()).decode()
     file_name = os.path.basename(file_path)
     ext = os.path.splitext(file_name)[1]  # get the file extension
@@ -234,13 +325,16 @@ 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")
 def read_file_content(file,max_length):
     if file.type == "application/json":
         content = json.load(file)
@@ -262,6 +356,8 @@ def read_file_content(file,max_length):
     else:
         return ""
 def chat_with_model(prompt, document_section, model_choice='gpt-3.5-turbo'):
     model = model_choice
     conversation = [{'role': 'system', 'content': 'You are a helpful assistant.'}]
@@ -290,6 +386,8 @@ def chat_with_model(prompt, document_section, model_choice='gpt-3.5-turbo'):
     st.write(time.time() - start_time)
     return full_reply_content
 def chat_with_file_contents(prompt, file_content, model_choice='gpt-3.5-turbo'):
     conversation = [{'role': 'system', 'content': 'You are a helpful assistant.'}]
     conversation.append({'role': 'user', 'content': prompt})
@@ -321,31 +419,34 @@ def extract_file_extension(file):
     else:
         raise ValueError(f"Unable to extract file extension from {file_name}")
 def pdf2txt(docs):
     text = ""
     for file in docs:
         file_extension = extract_file_extension(file)
         st.write(f"File type extension: {file_extension}")
-        try:
-            if file_extension.lower() in ['py', 'txt', 'html', 'htm', 'xml', 'json']:
-                text += file.getvalue().decode('utf-8')
-            elif file_extension.lower() == 'pdf':
-                from PyPDF2 import PdfReader
-                pdf = PdfReader(BytesIO(file.getvalue()))
-                for page in range(len(pdf.pages)):
-                    text += pdf.pages[page].extract_text() # new PyPDF2 syntax
-        except Exception as e:
-            st.write(f"Error processing file {file.name}: {e}")
     return text
 def txt2chunks(text):
     text_splitter = CharacterTextSplitter(separator="\n", chunk_size=1000, chunk_overlap=200, length_function=len)
     return text_splitter.split_text(text)
 def vector_store(text_chunks):
     embeddings = OpenAIEmbeddings(openai_api_key=key)
     return FAISS.from_texts(texts=text_chunks, embedding=embeddings)
 def get_chain(vectorstore):
     llm = ChatOpenAI()
     memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True)
@@ -378,13 +479,18 @@ def divide_prompt(prompt, max_length):
     chunks.append(' '.join(current_chunk))
     return chunks
 def create_zip_of_files(files):
     zip_name = "all_files.zip"
     with zipfile.ZipFile(zip_name, 'w') as zipf:
         for file in files:
             zipf.write(file)
     return zip_name
 def get_zip_download_link(zip_file):
     with open(zip_file, 'rb') as f:
         data = f.read()
@@ -392,13 +498,24 @@ def get_zip_download_link(zip_file):
     href = f'<a href="data:application/zip;base64,{b64}" download="{zip_file}">Download All</a>'
     return href
 API_URL_IE = f'https://tonpixzfvq3791u9.us-east-1.aws.endpoints.huggingface.cloud'
 headers = {
-	"Authorization": "Bearer XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX",
-	"Content-Type": "audio/wav"
 }
 def query(filename):
     with open(filename, "rb") as f:
         data = f.read()
@@ -412,7 +529,7 @@ 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:
 def save_and_play_audio(audio_recorder):
     audio_bytes = audio_recorder()
     if audio_bytes:
@@ -422,7 +539,7 @@ def save_and_play_audio(audio_recorder):
         st.audio(audio_bytes, format="audio/wav")
         return filename
-# 9B. Speech transcription to file output - OPENAI Whisper
 def transcribe_audio(filename):
     output = query(filename)
     return output
@@ -435,7 +552,11 @@ def whisper_main():
     filename = save_and_play_audio(audio_recorder)
     if filename is not None:
         transcription = transcribe_audio(filename)
-        transcription = transcription['text']
         st.write(transcription)
         response = StreamLLMChatResponse(transcription)
         # st.write(response) - redundant with streaming result?
@@ -443,6 +564,8 @@ def whisper_main():
         create_file(filename, transcription, response, should_save)
         #st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True)
 def main():
     st.title("AI Drome Llama")
@@ -461,14 +584,7 @@ def main():
     openai.api_key = os.getenv('OPENAI_KEY')
     menu = ["txt", "htm", "xlsx", "csv", "md", "py"]
     choice = st.sidebar.selectbox("Output File Type:", menu)
-    model_choice = st.sidebar.radio("Select Model:", ('gpt-3.5-turbo', 'gpt-3.5-turbo-0301'))
-    #filename = save_and_play_audio(audio_recorder)
-    #if filename is not None:
-    #    transcription = transcribe_audio(key, filename, "whisper-1")
-    #    st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True)
-    #    filename = None
     user_prompt = st.text_area("Enter prompts, instructions & questions:", '', height=100)
     collength, colupload = st.columns([2,3])  # adjust the ratio as needed
     with collength:
@@ -512,6 +628,8 @@ def main():
         filename = generate_filename(user_prompt, choice)
         create_file(filename, user_prompt, response, should_save)
         st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True)
     all_files = glob.glob("*.*")
     all_files = [file for file in all_files if len(os.path.splitext(file)[0]) >= 20]  # exclude files with short names
     all_files.sort(key=lambda x: (os.path.splitext(x)[1], x), reverse=True)  # sort by file type and file name in descending order
@@ -547,17 +665,32 @@ def main():
             if st.button("🗑", key="delete_"+file):
                 os.remove(file)
                 st.experimental_rerun()
     if len(file_contents) > 0:
         if next_action=='open':
             file_content_area = st.text_area("File Contents:", file_contents, height=500)
         if next_action=='md':
             st.markdown(file_contents)
         if next_action=='search':
             file_content_area = st.text_area("File Contents:", file_contents, height=500)
             st.write('Reasoning with your inputs...')
-            response = chat_with_model(user_prompt, file_contents, model_choice)
-            filename = generate_filename(file_contents, choice)
-            create_file(filename, user_prompt, response, should_save)
             st.experimental_rerun()
     # Feedback
@@ -588,6 +721,7 @@ def main():
                 filename = generate_filename(raw, 'txt')
                 create_file(filename, raw, '', should_save)
 if __name__ == "__main__":
     whisper_main()
-    main()

 from templates import bot_template, css, user_template
 from xml.etree import ElementTree as ET
+def add_Med_Licensing_Exam_Dataset():
+    import streamlit as st
+    from datasets import load_dataset
+    dataset = load_dataset("augtoma/usmle_step_1")['test']  # Using 'test' split
+    st.title("USMLE Step 1 Dataset Viewer")
+    if len(dataset) == 0:
+        st.write("😢 The dataset is empty.")
+    else:
+        st.write("""
+        🔍 Use the search box to filter questions or use the grid to scroll through the dataset.
+        """)
+        # 👩‍🔬 Search Box
+        search_term = st.text_input("Search for a specific question:", "")
+        # 🎛 Pagination
+        records_per_page = 100
+        num_records = len(dataset)
+        num_pages = max(int(num_records / records_per_page), 1)
+        # Skip generating the slider if num_pages is 1 (i.e., all records fit in one page)
+        if num_pages > 1:
+            page_number = st.select_slider("Select page:", options=list(range(1, num_pages + 1)))
+        else:
+            page_number = 1  # Only one page
+        # 📊 Display Data
+        start_idx = (page_number - 1) * records_per_page
+        end_idx = start_idx + records_per_page
+        # 🧪 Apply the Search Filter
+        filtered_data = []
+        for record in dataset[start_idx:end_idx]:
+            if isinstance(record, dict) and 'text' in record and 'id' in record:
+                if search_term:
+                    if search_term.lower() in record['text'].lower():
+                        filtered_data.append(record)
+                else:
+                    filtered_data.append(record)
+        # 🌐 Render the Grid
+        for record in filtered_data:
+            st.write(f"## Question ID: {record['id']}")
+            st.write(f"### Question:")
+            st.write(f"{record['text']}")
+            st.write(f"### Answer:")
+            st.write(f"{record['answer']}")
+            st.write("---")
+        st.write(f"😊 Total Records: {num_records} | 📄 Displaying {start_idx+1} to {min(end_idx, num_records)}")
+# 1. Constants and Top Level UI Variables
+# My Inference API Copy
+# API_URL = 'https://qe55p8afio98s0u3.us-east-1.aws.endpoints.huggingface.cloud'  # Dr Llama
+# Original:
+API_URL = "https://api-inference.huggingface.co/models/meta-llama/Llama-2-7b-chat-hf"
 API_KEY = os.getenv('API_KEY')
+MODEL1="meta-llama/Llama-2-7b-chat-hf"
+MODEL1URL="https://huggingface.co/meta-llama/Llama-2-7b-chat-hf"
+HF_KEY = os.getenv('HF_KEY')
 headers = {
+    "Authorization": f"Bearer {HF_KEY}",
     "Content-Type": "application/json"
 }
 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")
 should_save = st.sidebar.checkbox("💾 Save", value=True, help="Save your session data.")
+# 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
         if col7[0].button("More Funny Rhymes 🎙️"):
             StreamLLMChatResponse(descriptions["More Funny Rhymes 🎙️"])
+def addDocumentHTML5(result):
+    documentHTML5='''
+    <!DOCTYPE html>
+    <html>
+    <head>
+        <title>Read It Aloud</title>
+        <script type="text/javascript">
+            function readAloud() {
+                const text = document.getElementById("textArea").value;
+                const speech = new SpeechSynthesisUtterance(text);
+                window.speechSynthesis.speak(speech);
+            }
+        </script>
+    </head>
+    <body>
+        <h1>🔊 Read It Aloud</h1>
+        <textarea id="textArea" rows="10" cols="80">
+    '''
+    documentHTML5 = documentHTML5 + result
+    documentHTML5 = documentHTML5 + '''
+        </textarea>
+        <br>
+        <button onclick="readAloud()">🔊 Read Aloud</button>
+    </body>
+    </html>
+    '''
+    import streamlit.components.v1 as components  # Import Streamlit
+    components.html(documentHTML5, width=1280, height=1024)
+    return result
+# 3. Stream Llama Response
+# @st.cache_resource
 def StreamLLMChatResponse(prompt):
     try:
             except:
                 st.write('Stream llm issue')
+        add_documentHTML5(result)
     except:
+        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
 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 == "_")[:45]
     return f"{safe_date_time}_{safe_prompt}.{file_type}"
+# 6. Speech transcription via OpenAI service
 def transcribe_audio(openai_key, file_path, model):
     openai.api_key = openai_key
     OPENAI_API_URL = "https://api.openai.com/v1/audio/transcriptions"
         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:
         return filename
     return None
+# 8. File creator that interprets type and creates output file for text, markdown and code
 def create_file(filename, prompt, response, should_save=True):
     if not should_save:
         return
     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}.md", 'w') as file:
+            content = prompt.strip() + '\r\n' + response
+            file.write(content)
         if has_python_code:
             python_code = re.findall(r"```python([\s\S]*?)```", response)[0].strip()
             with open(f"{base_filename}-Code.py", 'w') as file:
                 file.write(python_code)
+            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:
+        data = file.read()
     b64 = base64.b64encode(data.encode()).decode()
     file_name = os.path.basename(file_path)
     ext = os.path.splitext(file_name)[1]  # get the file extension
     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":
         content = json.load(file)
     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
     conversation = [{'role': 'system', 'content': 'You are a helpful assistant.'}]
     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.'}]
     conversation.append({'role': 'user', 'content': prompt})
     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 = ""
     for file in docs:
         file_extension = extract_file_extension(file)
         st.write(f"File type extension: {file_extension}")
+        if file_extension.lower() in ['py', 'txt', 'html', 'htm', 'xml', 'json']:
+            text += file.getvalue().decode('utf-8')
+        elif file_extension.lower() == 'pdf':
+            from PyPDF2 import PdfReader
+            pdf = PdfReader(BytesIO(file.getvalue()))
+            for page in range(len(pdf.pages)):
+                text += pdf.pages[page].extract_text() # new PyPDF2 syntax
     return text
 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()
     memory = ConversationBufferMemory(memory_key='chat_history', return_messages=True)
     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"
     with zipfile.ZipFile(zip_name, 'w') as zipf:
         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:
         data = f.read()
     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
+# My Inference Endpoint
 API_URL_IE = f'https://tonpixzfvq3791u9.us-east-1.aws.endpoints.huggingface.cloud'
+# Original
+API_URL_IE = "https://api-inference.huggingface.co/models/openai/whisper-small.en"
+MODEL2 = "openai/whisper-small.en"
+MODEL2_URL = "https://huggingface.co/openai/whisper-small.en"
+#headers = {
+#	"Authorization": "Bearer XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX",
+#	"Content-Type": "audio/wav"
+#}
+HF_KEY = os.getenv('HF_KEY')
 headers = {
+    "Authorization": f"Bearer {HF_KEY}",
+    "Content-Type": "audio/wav"
 }
+#@st.cache_resource
 def query(filename):
     with open(filename, "rb") as f:
         data = f.read()
     safe_prompt = "".join(x for x in replaced_prompt if x.isalnum() or x == "_")[:90]
     return f"{safe_date_time}_{safe_prompt}.{file_type}"
+# 15. Audio recorder to Wav file
 def save_and_play_audio(audio_recorder):
     audio_bytes = audio_recorder()
     if audio_bytes:
         st.audio(audio_bytes, format="audio/wav")
         return filename
+# 16. Speech transcription to file output
 def transcribe_audio(filename):
     output = query(filename)
     return output
     filename = save_and_play_audio(audio_recorder)
     if filename is not None:
         transcription = transcribe_audio(filename)
+        try:
+            transcription = transcription['text']
+        except:
+            st.write('Whisper model is asleep. Starting up now on T4 GPU - please give 5 minutes then retry as it scales up from zero to activate running container(s).')
         st.write(transcription)
         response = StreamLLMChatResponse(transcription)
         # st.write(response) - redundant with streaming result?
         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")
     openai.api_key = os.getenv('OPENAI_KEY')
     menu = ["txt", "htm", "xlsx", "csv", "md", "py"]
     choice = st.sidebar.selectbox("Output File Type:", menu)
+    model_choice = st.sidebar.radio("Select Model:", ('gpt-3.5-turbo', 'gpt-3.5-turbo-0301'))
     user_prompt = st.text_area("Enter prompts, instructions & questions:", '', height=100)
     collength, colupload = st.columns([2,3])  # adjust the ratio as needed
     with collength:
         filename = generate_filename(user_prompt, choice)
         create_file(filename, user_prompt, response, should_save)
         st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True)
+    # Compose a file sidebar of past encounters
     all_files = glob.glob("*.*")
     all_files = [file for file in all_files if len(os.path.splitext(file)[0]) >= 20]  # exclude files with short names
     all_files.sort(key=lambda x: (os.path.splitext(x)[1], x), reverse=True)  # sort by file type and file name in descending order
             if st.button("🗑", key="delete_"+file):
                 os.remove(file)
                 st.experimental_rerun()
     if len(file_contents) > 0:
         if next_action=='open':
             file_content_area = st.text_area("File Contents:", file_contents, height=500)
+            #addDocumentHTML5(file_contents)
         if next_action=='md':
             st.markdown(file_contents)
+            #addDocumentHTML5(file_contents)
         if next_action=='search':
             file_content_area = st.text_area("File Contents:", file_contents, height=500)
             st.write('Reasoning with your inputs...')
+            # new - llama
+            response = StreamLLMChatResponse(file_contents)
+            filename = generate_filename(user_prompt, ".md")
+            create_file(filename, file_contents, response, should_save)
+            #addDocumentHTML5(file_contents)
+            addDocumentHTML5(response)
+            # old - gpt
+            #response = chat_with_model(user_prompt, file_contents, model_choice)
+            #filename = generate_filename(file_contents, choice)
+            #create_file(filename, user_prompt, response, should_save)
             st.experimental_rerun()
     # Feedback
                 filename = generate_filename(raw, 'txt')
                 create_file(filename, raw, '', should_save)
+# 18. Run AI Pipeline
 if __name__ == "__main__":
     whisper_main()
+    main()