import huggingface_hub from huggingface_hub import InferenceClient import streamlit as st import streamlit.components.v1 as components import openai import os import base64 import glob import io import json import mistune import pytz import math import requests import sys import time import re import textract import zipfile import random from datetime import datetime from openai import ChatCompletion from xml.etree import ElementTree as ET from bs4 import BeautifulSoup from collections import deque from audio_recorder_streamlit import audio_recorder from dotenv import load_dotenv from PyPDF2 import PdfReader from langchain.text_splitter import CharacterTextSplitter from langchain.embeddings import OpenAIEmbeddings from langchain.vectorstores import FAISS from langchain.chat_models import ChatOpenAI from langchain.memory import ConversationBufferMemory from langchain.chains import ConversationalRetrievalChain from templates import css, bot_template, user_template from io import BytesIO import streamlit.components.v1 as components # Import Streamlit Components for HTML5 # page config and sidebar declares up front allow all other functions to see global class variables st.set_page_config(page_title="AI Human Body - Homunculus Body Reasoner", layout="wide") should_save = st.sidebar.checkbox("๐Ÿ’พ Save", value=True) col1, col2, col3, col4 = st.columns(4) with col1: with st.expander("Settings ๐Ÿง ๐Ÿ’พ", expanded=True): # File type for output, model choice 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')) # Define a context dictionary to maintain the state between exec calls context = {} def SpeechSynthesis(result): documentHTML5=''' Read It Aloud

๐Ÿ”Š Read It Aloud


''' components.html(documentHTML5, width=1280, height=1024) #return result def create_file(filename, prompt, response, should_save=True): if not should_save: return # Extract base filename without extension base_filename, ext = os.path.splitext(filename) # Initialize the combined content combined_content = "" # Add Prompt with markdown title and emoji combined_content += "# Prompt ๐Ÿ“\n" + prompt + "\n\n" # Add Response with markdown title and emoji combined_content += "# Response ๐Ÿ’ฌ\n" + response + "\n\n" # Check for code blocks in the response resources = re.findall(r"```([\s\S]*?)```", response) for resource in resources: # Check if the resource contains Python code if "python" in resource.lower(): # Remove the 'python' keyword from the code block cleaned_code = re.sub(r'^\s*python', '', resource, flags=re.IGNORECASE | re.MULTILINE) # Add Code Results title with markdown and emoji combined_content += "# Code Results ๐Ÿš€\n" # Redirect standard output to capture it original_stdout = sys.stdout sys.stdout = io.StringIO() # Execute the cleaned Python code within the context try: exec(cleaned_code, context) code_output = sys.stdout.getvalue() combined_content += f"```\n{code_output}\n```\n\n" realtimeEvalResponse = "# Code Results ๐Ÿš€\n" + "```" + code_output + "```\n\n" st.write(realtimeEvalResponse) except Exception as e: combined_content += f"```python\nError executing Python code: {e}\n```\n\n" # Restore the original standard output sys.stdout = original_stdout else: # Add non-Python resources with markdown and emoji combined_content += "# Resource ๐Ÿ› ๏ธ\n" + "```" + resource + "```\n\n" # Save the combined content to a Markdown file if should_save: with open(f"{base_filename}.md", 'w') as file: file.write(combined_content) # Read it aloud def readitaloud(result): documentHTML5=''' Read It Aloud

๐Ÿ”Š Read It Aloud


''' components.html(documentHTML5, width=800, height=300) #return result 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}" # 3. Stream Llama Response # @st.cache_resource def StreamLLMChatResponse(prompt): # My Inference API Copy API_URL = 'https://qe55p8afio98s0u3.us-east-1.aws.endpoints.huggingface.cloud' # Dr Llama API_KEY = os.getenv('API_KEY') #try: endpoint_url = API_URL hf_token = API_KEY client = InferenceClient(endpoint_url, token=hf_token) gen_kwargs = dict( max_new_tokens=512, top_k=30, top_p=0.9, temperature=0.2, repetition_penalty=1.02, stop_sequences=["\nUser:", "<|endoftext|>", ""], ) stream = client.text_generation(prompt, stream=True, details=True, **gen_kwargs) report=[] res_box = st.empty() collected_chunks=[] collected_messages=[] allresults='' for r in stream: if r.token.special: continue if r.token.text in gen_kwargs["stop_sequences"]: break collected_chunks.append(r.token.text) chunk_message = r.token.text collected_messages.append(chunk_message) #try: report.append(r.token.text) if len(r.token.text) > 0: result="".join(report).strip() res_box.markdown(f'*{result}*') #except: #st.write('Stream llm issue') SpeechSynthesis(result) return 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).') # Chat and Chat with files 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.'}] conversation.append({'role': 'user', 'content': prompt}) if len(document_section)>0: conversation.append({'role': 'assistant', 'content': document_section}) start_time = time.time() report = [] res_box = st.empty() collected_chunks = [] collected_messages = [] key = os.getenv('OPENAI_API_KEY') openai.api_key = key for chunk in openai.ChatCompletion.create( model='gpt-3.5-turbo', messages=conversation, temperature=0.5, stream=True ): collected_chunks.append(chunk) # save the event response chunk_message = chunk['choices'][0]['delta'] # extract the message collected_messages.append(chunk_message) # save the message content=chunk["choices"][0].get("delta",{}).get("content") try: report.append(content) if len(content) > 0: result = "".join(report).strip() res_box.markdown(f'*{result}*') except: st.write(' ') full_reply_content = ''.join([m.get('content', '') for m in collected_messages]) st.write("Elapsed time:") st.write(time.time() - start_time) readitaloud(full_reply_content) filename = generate_filename(full_reply_content, choice) create_file(filename, prompt, full_reply_content, should_save) 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}) if len(file_content)>0: conversation.append({'role': 'assistant', 'content': file_content}) response = openai.ChatCompletion.create(model=model_choice, messages=conversation) return response['choices'][0]['message']['content'] def link_button_with_emoji(url, title, emoji_summary): emojis = ["๐Ÿ’‰", "๐Ÿฅ", "๐ŸŒก๏ธ", "๐Ÿฉบ", "๐Ÿ”ฌ", "๐Ÿ’Š", "๐Ÿงช", "๐Ÿ‘จโ€โš•๏ธ", "๐Ÿ‘ฉโ€โš•๏ธ"] random_emoji = random.choice(emojis) st.markdown(f"[{random_emoji} {emoji_summary} - {title}]({url})") # Homunculus parts and their corresponding emojis homunculus_parts = { "Head": "๐Ÿง ", "Brain": "๐Ÿง ", "Left Eye": "๐Ÿ‘๏ธ", "Right Eye": "๐Ÿ‘๏ธ", "Left Eyebrow": "๐Ÿคจ", "Right Eyebrow": "๐Ÿคจ", "Nose": "๐Ÿ‘ƒ", "Mouth": "๐Ÿ‘„", "Neck": "๐Ÿงฃ", "Left Shoulder": "๐Ÿ’ช", "Right Shoulder": "๐Ÿ’ช", "Left Upper Arm": "๐Ÿ’ช", "Right Upper Arm": "๐Ÿ’ช", "Left Elbow": "๐Ÿ’ช", "Right Elbow": "๐Ÿ’ช", "Left Forearm": "๐Ÿ’ช", "Right Forearm": "๐Ÿ’ช", "Left Wrist": "โœŠ", "Right Wrist": "โœŠ", "Left Hand": "๐Ÿคฒ", "Right Hand": "๐Ÿคฒ", "Chest": "๐Ÿ‘•", "Abdomen": "๐Ÿ‘•", "Pelvis": "๐Ÿฉฒ", "Left Hip": "๐Ÿฆต", "Right Hip": "๐Ÿฆต", "Left Thigh": "๐Ÿฆต", "Right Thigh": "๐Ÿฆต", "Left Knee": "๐Ÿฆต", "Right Knee": "๐Ÿฆต", "Left Shin": "๐Ÿฆต", "Right Shin": "๐Ÿฆต" } homunculus_parts_extended = { "Head": "๐Ÿง  (Center of Thought and Control)", "Brain": "๐Ÿง  (Organ of Intelligence and Processing)", "Left Eye": "๐Ÿ‘๏ธ (Vision and Perception - Left)", "Right Eye": "๐Ÿ‘๏ธ (Vision and Perception - Right)", "Left Eyebrow": "๐Ÿคจ (Facial Expression - Left Eyebrow)", "Right Eyebrow": "๐Ÿคจ (Facial Expression - Right Eyebrow)", "Nose": "๐Ÿ‘ƒ (Smell and Breathing)", "Mouth": "๐Ÿ‘„ (Speech and Eating)", "Neck": "๐Ÿงฃ (Support and Movement of Head)", "Left Shoulder": "๐Ÿ’ช (Arm Movement and Strength - Left)", "Right Shoulder": "๐Ÿ’ช (Arm Movement and Strength - Right)", "Left Upper Arm": "๐Ÿ’ช (Support and Lifting - Left Upper)", "Right Upper Arm": "๐Ÿ’ช (Support and Lifting - Right Upper)", "Left Elbow": "๐Ÿ’ช (Arm Bending and Flexing - Left)", "Right Elbow": "๐Ÿ’ช (Arm Bending and Flexing - Right)", "Left Forearm": "๐Ÿ’ช (Wrist and Hand Movement - Left)", "Right Forearm": "๐Ÿ’ช (Wrist and Hand Movement - Right)", "Left Wrist": "โœŠ (Hand Articulation and Rotation - Left)", "Right Wrist": "โœŠ (Hand Articulation and Rotation - Right)", "Left Hand": "๐Ÿคฒ (Grasping and Touch - Left)", "Right Hand": "๐Ÿคฒ (Grasping and Touch - Right)", "Chest": "๐Ÿ‘• (Protection of Heart and Lungs)", "Abdomen": "๐Ÿ‘• (Digestive Organs and Processing)", "Pelvis": "๐Ÿฉฒ (Support for Lower Limbs and Organs)", "Left Hip": "๐Ÿฆต (Support and Movement - Left Hip)", "Right Hip": "๐Ÿฆต (Support and Movement - Right Hip)", "Left Thigh": "๐Ÿฆต (Support and Movement - Left Thigh)", "Right Thigh": "๐Ÿฆต (Support and Movement - Right Thigh)", "Left Knee": "๐Ÿฆต (Leg Bending and Flexing - Left)", "Right Knee": "๐Ÿฆต (Leg Bending and Flexing - Right)", "Left Shin": "๐Ÿฆต (Lower Leg Support - Left)", "Right Shin": "๐Ÿฆต (Lower Leg Support - Right)", "Left Foot": "๐Ÿฆถ (Support, Balance, and Locomotion - Left)", "Right Foot": "๐Ÿฆถ (Support, Balance, and Locomotion - Right)" } # Function to display the homunculus parts with expanders and chat buttons def display_homunculus_parts(): st.title("Homunculus Model") with st.expander(f"Head ({homunculus_parts_extended['Head']})", expanded=False): head_parts = ["Left Eye", "Right Eye", "Left Eyebrow", "Right Eyebrow", "Nose", "Mouth"] for part in head_parts: # Extracting the function/description from the extended dictionary part_description = homunculus_parts_extended[part].split('(')[1].rstrip(')') prompt = f"Learn about the key features and functions of the {part} - {part_description}" if st.button(f"Explore {part}", key=part): #response = chat_with_model(prompt, part) # GPT response = StreamLLMChatResponse(prompt) # Llama with st.expander(f"Brain ({homunculus_parts['Brain']})", expanded=False): brain_parts = { "Neocortex": "๐ŸŒ€ - Involved in higher-order brain functions such as sensory perception, cognition, and spatial reasoning.", "Limbic System": "โค๏ธ - Supports functions including emotion, behavior, motivation, long-term memory, and olfaction.", "Brainstem": "๐ŸŒฑ - Controls basic body functions like breathing, swallowing, heart rate, blood pressure, and consciousness.", "Cerebellum": "๐Ÿงฉ - Coordinates voluntary movements like posture, balance, and speech, resulting in smooth muscular activity.", "Thalamus": "๐Ÿ”” - Channels sensory and motor signals to the cerebral cortex, and regulates consciousness and sleep.", "Hypothalamus": "๐ŸŒก๏ธ - Controls body temperature, hunger, thirst, fatigue, and circadian cycles.", "Hippocampus": "๐Ÿš - Essential for the formation of new memories and associated with learning and emotions.", "Frontal Lobe": "๐Ÿ’ก - Associated with decision making, problem solving, and planning.", "Parietal Lobe": "๐Ÿคš - Processes sensory information it receives from the outside world, mainly relating to spatial sense and navigation.", "Temporal Lobe": "๐Ÿ‘‚ - Involved in processing auditory information and is also important for the processing of semantics in both speech and vision.", "Occipital Lobe": "๐Ÿ‘๏ธ - Main center for visual processing." } for part, description in brain_parts.items(): # Formatting the prompt in markdown style for enhanced learning prompt = f"Create a markdown outline with emojis to explain the {part} and its role in the brain: {description}" if st.button(f"Explore {part} ๐Ÿง ", key=part): #response = chat_with_model(prompt, part) response = StreamLLMChatResponse(prompt) # Llama # Displaying central body parts central_parts = ["Neck", "Chest", "Abdomen", "Pelvis"] for part in central_parts: with st.expander(f"{part} ({homunculus_parts_extended[part]})", expanded=False): prompt = f"Learn about the key features and functions of the {part} - {homunculus_parts_extended[part].split(' ')[-1]}" if st.button(f"Explore {part} ๐Ÿงฃ", key=part): #response = chat_with_model(prompt, part) response = StreamLLMChatResponse(prompt) # Llama # Displaying symmetric body parts symmetric_parts = ["Shoulder", "Upper Arm", "Elbow", "Forearm", "Wrist", "Hand", "Hip", "Thigh", "Knee", "Shin", "Foot"] for part in symmetric_parts: col1, col2 = st.columns(2) with col1: with st.expander(f"Left {part} ({homunculus_parts_extended[f'Left {part}']})", expanded=False): prompt = f"Learn about the key features and functions of the Left {part} - {homunculus_parts_extended[f'Left {part}'].split(' ')[-1]}" if st.button(f"Explore Left {part} ๐Ÿ’ช", key=f"Left {part}"): #response = chat_with_model(prompt, f"Left {part}") response = StreamLLMChatResponse(prompt) # Llama with col2: with st.expander(f"Right {part} ({homunculus_parts_extended[f'Right {part}']})", expanded=False): prompt = f"Learn about the key features and functions of the Right {part} - {homunculus_parts_extended[f'Right {part}'].split(' ')[-1]}" if st.button(f"Explore Right {part} ๐Ÿ’ช", key=f"Right {part}"): #response = chat_with_model(prompt, f"Right {part}") response = StreamLLMChatResponse(prompt) # Llama # Define function to add paper buttons and links def add_paper_buttons_and_links(): # Homunculus page = st.sidebar.radio("Choose a page:", ["Detailed Homunculus Model"]) if page == "Detailed Homunculus Model": display_homunculus_parts() col1, col2, col3, col4 = st.columns(4) with col1: with st.expander("MemGPT ๐Ÿง ๐Ÿ’พ", expanded=False): link_button_with_emoji("https://arxiv.org/abs/2310.08560", "MemGPT", "๐Ÿง ๐Ÿ’พ Memory OS") outline_memgpt = "Memory Hierarchy, Context Paging, Self-directed Memory Updates, Memory Editing, Memory Retrieval, Preprompt Instructions, Semantic Memory, Episodic Memory, Emotional Contextual Understanding" if st.button("Discuss MemGPT Features"): prompt = "Discuss the key features of MemGPT: " + outline_memgpt #chat_with_model(prompt, "MemGPT") response = StreamLLMChatResponse(prompt) # Llama with col2: with st.expander("AutoGen ๐Ÿค–๐Ÿ”—", expanded=False): link_button_with_emoji("https://arxiv.org/abs/2308.08155", "AutoGen", "๐Ÿค–๐Ÿ”— Multi-Agent LLM") outline_autogen = "Cooperative Conversations, Combining Capabilities, Complex Task Solving, Divergent Thinking, Factuality, Highly Capable Agents, Generic Abstraction, Effective Implementation" if st.button("Explore AutoGen Multi-Agent LLM"): prompt = "Explore the key features of AutoGen: " + outline_autogen #chat_with_model(prompt, "AutoGen") response = StreamLLMChatResponse(prompt) # Llama with col3: with st.expander("Whisper ๐Ÿ”Š๐Ÿง‘โ€๐Ÿš€", expanded=False): link_button_with_emoji("https://arxiv.org/abs/2212.04356", "Whisper", "๐Ÿ”Š๐Ÿง‘โ€๐Ÿš€ Robust STT") outline_whisper = "Scaling, Deep Learning Approaches, Weak Supervision, Zero-shot Transfer Learning, Accuracy & Robustness, Pre-training Techniques, Broad Range of Environments, Combining Multiple Datasets" if st.button("Learn About Whisper STT"): prompt = "Learn about the key features of Whisper: " + outline_whisper #chat_with_model(prompt, "Whisper") response = StreamLLMChatResponse(prompt) # Llama with col4: with st.expander("ChatDev ๐Ÿ’ฌ๐Ÿ’ป", expanded=False): link_button_with_emoji("https://arxiv.org/pdf/2307.07924.pdf", "ChatDev", "๐Ÿ’ฌ๐Ÿ’ป Comm. Agents") outline_chatdev = "Effective Communication, Comprehensive Software Solutions, Diverse Social Identities, Tailored Codes, Environment Dependencies, User Manuals" if st.button("Deep Dive into ChatDev"): prompt = "Deep dive into the features of ChatDev: " + outline_chatdev #chat_with_model(prompt, "ChatDev") response = StreamLLMChatResponse(prompt) # Llama add_paper_buttons_and_links() # Process user input is a post processor algorithm which runs after document embedding vector DB play of GPT on context of documents.. def process_user_input(user_question): # Check and initialize 'conversation' in session state if not present if 'conversation' not in st.session_state: st.session_state.conversation = {} # Initialize with an empty dictionary or an appropriate default value response = st.session_state.conversation({'question': user_question}) st.session_state.chat_history = response['chat_history'] for i, message in enumerate(st.session_state.chat_history): template = user_template if i % 2 == 0 else bot_template st.write(template.replace("{{MSG}}", message.content), unsafe_allow_html=True) # Save file output from PDF query results filename = generate_filename(user_question, 'txt') create_file(filename, user_question, message.content, should_save) # New functionality to create expanders and buttons create_expanders_and_buttons(message.content) def create_expanders_and_buttons(content): # Split the content into paragraphs paragraphs = content.split("\n\n") for paragraph in paragraphs: # Identify the header and detail in the paragraph header, detail = extract_feature_and_detail(paragraph) if header and detail: with st.expander(header, expanded=False): if st.button(f"Explore {header}"): expanded_outline = "Expand on the feature: " + detail #chat_with_model(expanded_outline, header) response = StreamLLMChatResponse(expanded_outline) # Llama def extract_feature_and_detail(paragraph): # Use regex to find the header and detail in the paragraph match = re.match(r"(.*?):(.*)", paragraph) if match: header = match.group(1).strip() detail = match.group(2).strip() return header, detail return None, None def transcribe_audio(file_path, model): key = os.getenv('OPENAI_API_KEY') headers = { "Authorization": f"Bearer {key}", } with open(file_path, 'rb') as f: data = {'file': f} st.write("Read file {file_path}", file_path) OPENAI_API_URL = "https://api.openai.com/v1/audio/transcriptions" response = requests.post(OPENAI_API_URL, headers=headers, files=data, data={'model': model}) if response.status_code == 200: st.write(response.json()) prompt = response.json().get('text') chatResponse = chat_with_model(prompt, '') # ************************************* response = StreamLLMChatResponse(prompt) # Llama transcript = response.json().get('text') #st.write('Responses:') #st.write(chatResponse) filename = generate_filename(transcript, 'txt') #create_file(filename, transcript, chatResponse) response = chatResponse user_prompt = transcript create_file(filename, user_prompt, response, should_save) return transcript else: st.write(response.json()) st.error("Error in API call.") return None def save_and_play_audio(audio_recorder): audio_bytes = audio_recorder() if audio_bytes: filename = generate_filename("Recording", "wav") with open(filename, 'wb') as f: f.write(audio_bytes) st.audio(audio_bytes, format="audio/wav") return filename return None 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 if ext == '.txt': mime_type = 'text/plain' elif ext == '.py': mime_type = 'text/plain' elif ext == '.xlsx': mime_type = 'text/plain' elif ext == '.csv': mime_type = 'text/plain' elif ext == '.htm': mime_type = 'text/html' elif ext == '.md': mime_type = 'text/markdown' else: mime_type = 'application/octet-stream' # general binary data type href = f'{file_name}' 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) return str(content) elif file.type == "text/html" or file.type == "text/htm": content = BeautifulSoup(file, "html.parser") return content.text elif file.type == "application/xml" or file.type == "text/xml": tree = ET.parse(file) root = tree.getroot() xml = CompressXML(ET.tostring(root, encoding='unicode')) return xml elif file.type == "text/markdown" or file.type == "text/md": md = mistune.create_markdown() content = md(file.read().decode()) return content elif file.type == "text/plain": return file.getvalue().decode() else: return "" def extract_mime_type(file): # Check if the input is a string if isinstance(file, str): pattern = r"type='(.*?)'" match = re.search(pattern, file) if match: return match.group(1) else: raise ValueError(f"Unable to extract MIME type from {file}") # If it's not a string, assume it's a streamlit.UploadedFile object elif isinstance(file, streamlit.UploadedFile): return file.type else: raise TypeError("Input should be a string or a streamlit.UploadedFile object") def extract_file_extension(file): # get the file name directly from the UploadedFile object file_name = file.name pattern = r".*?\.(.*?)$" match = re.search(pattern, file_name) if match: return match.group(1) 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) # print the file extension st.write(f"File type extension: {file_extension}") # read the file according to its 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): key = os.getenv('OPENAI_API_KEY') 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) return ConversationalRetrievalChain.from_llm(llm=llm, retriever=vectorstore.as_retriever(), memory=memory) def divide_prompt(prompt, max_length): words = prompt.split() chunks = [] current_chunk = [] current_length = 0 for word in words: if len(word) + current_length <= max_length: current_length += len(word) + 1 # Adding 1 to account for spaces current_chunk.append(word) else: chunks.append(' '.join(current_chunk)) current_chunk = [word] current_length = len(word) chunks.append(' '.join(current_chunk)) # Append the final chunk return chunks def create_zip_of_files(files): """ Create a zip file from a list of 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): """ Generate a link to download the zip file. """ with open(zip_file, 'rb') as f: data = f.read() b64 = base64.b64encode(data).decode() href = f'Download All' return href def main(): # Audio, transcribe, GPT: filename = save_and_play_audio(audio_recorder) if filename is not None: try: transcription = transcribe_audio(filename, "whisper-1") except: st.write(' ') st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True) filename = None # prompt interfaces user_prompt = st.text_area("Enter prompts, instructions & questions:", '', height=100) # file section interface for prompts against large documents as context collength, colupload = st.columns([2,3]) # adjust the ratio as needed with collength: max_length = st.slider("File section length for large files", min_value=1000, max_value=128000, value=12000, step=1000) with colupload: uploaded_file = st.file_uploader("Add a file for context:", type=["pdf", "xml", "json", "xlsx", "csv", "html", "htm", "md", "txt"]) # Document section chat document_sections = deque() document_responses = {} if uploaded_file is not None: file_content = read_file_content(uploaded_file, max_length) document_sections.extend(divide_document(file_content, max_length)) if len(document_sections) > 0: if st.button("๐Ÿ‘๏ธ View Upload"): st.markdown("**Sections of the uploaded file:**") for i, section in enumerate(list(document_sections)): st.markdown(f"**Section {i+1}**\n{section}") st.markdown("**Chat with the model:**") for i, section in enumerate(list(document_sections)): if i in document_responses: st.markdown(f"**Section {i+1}**\n{document_responses[i]}") else: if st.button(f"Chat about Section {i+1}"): st.write('Reasoning with your inputs...') #response = chat_with_model(user_prompt, section, model_choice) response = StreamLLMChatResponse(user_prompt + ' ' + section) # Llama st.write('Response:') st.write(response) document_responses[i] = response filename = generate_filename(f"{user_prompt}_section_{i+1}", choice) create_file(filename, user_prompt, response, should_save) st.sidebar.markdown(get_table_download_link(filename), unsafe_allow_html=True) if st.button('๐Ÿ’ฌ Chat'): st.write('Reasoning with your inputs...') # Divide the user_prompt into smaller sections user_prompt_sections = divide_prompt(user_prompt, max_length) full_response = '' for prompt_section in user_prompt_sections: # Process each section with the model #response = chat_with_model(prompt_section, ''.join(list(document_sections)), model_choice) response = StreamLLMChatResponse(prompt_section + ''.join(list(document_sections))) # Llama full_response += response + '\n' # Combine the responses response = full_response st.write('Response:') st.write(response) 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 # Sidebar buttons Download All and Delete All colDownloadAll, colDeleteAll = st.sidebar.columns([3,3]) with colDownloadAll: if st.button("โฌ‡๏ธ Download All"): zip_file = create_zip_of_files(all_files) st.markdown(get_zip_download_link(zip_file), unsafe_allow_html=True) with colDeleteAll: if st.button("๐Ÿ—‘ Delete All"): for file in all_files: os.remove(file) st.experimental_rerun() # Sidebar of Files Saving History and surfacing files as context of prompts and responses file_contents='' next_action='' for file in all_files: col1, col2, col3, col4, col5 = st.sidebar.columns([1,6,1,1,1]) # adjust the ratio as needed with col1: if st.button("๐ŸŒ", key="md_"+file): # md emoji button with open(file, 'r') as f: file_contents = f.read() next_action='md' with col2: st.markdown(get_table_download_link(file), unsafe_allow_html=True) with col3: if st.button("๐Ÿ“‚", key="open_"+file): # open emoji button with open(file, 'r') as f: file_contents = f.read() next_action='open' with col4: if st.button("๐Ÿ”", key="read_"+file): # search emoji button with open(file, 'r') as f: file_contents = f.read() next_action='search' with col5: if st.button("๐Ÿ—‘", 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) response = StreamLLMChatResponse(user_prompt + ' ' + file_contents) # Llama filename = generate_filename(file_contents, choice) create_file(filename, user_prompt, response, should_save) st.experimental_rerun() if __name__ == "__main__": main() load_dotenv() st.write(css, unsafe_allow_html=True) st.header("Chat with documents :books:") user_question = st.text_input("Ask a question about your documents:") if user_question: process_user_input(user_question) with st.sidebar: st.subheader("Your documents") docs = st.file_uploader("import documents", accept_multiple_files=True) with st.spinner("Processing"): raw = pdf2txt(docs) if len(raw) > 0: length = str(len(raw)) text_chunks = txt2chunks(raw) vectorstore = vector_store(text_chunks) st.session_state.conversation = get_chain(vectorstore) st.markdown('# AI Search Index of Length:' + length + ' Created.') # add timing filename = generate_filename(raw, 'txt') create_file(filename, raw, '', should_save)