| import requests |
| import os, sys, json |
| import gradio as gr |
| import openai |
| from openai import OpenAI |
| import time |
| import re |
| import io |
| from PIL import Image, ImageDraw, ImageOps, ImageFont |
| import base64 |
|
|
| from tavily import TavilyClient |
|
|
| from langchain.chains import LLMChain, RetrievalQA |
| from langchain.chat_models import ChatOpenAI |
| from langchain.document_loaders import PyPDFLoader, WebBaseLoader, UnstructuredWordDocumentLoader, DirectoryLoader |
| from langchain.document_loaders.blob_loaders.youtube_audio import YoutubeAudioLoader |
| from langchain.document_loaders.generic import GenericLoader |
| from langchain.document_loaders.parsers import OpenAIWhisperParser |
| from langchain.schema import AIMessage, HumanMessage |
| from langchain.llms import HuggingFaceHub |
| from langchain.llms import HuggingFaceTextGenInference |
| from langchain.embeddings import HuggingFaceInstructEmbeddings, HuggingFaceEmbeddings, HuggingFaceBgeEmbeddings, HuggingFaceInferenceAPIEmbeddings |
| from langchain.retrievers.tavily_search_api import TavilySearchAPIRetriever |
|
|
| from langchain.embeddings.openai import OpenAIEmbeddings |
| from langchain.prompts import PromptTemplate |
| from langchain.text_splitter import RecursiveCharacterTextSplitter |
| from langchain.vectorstores import Chroma |
| from chromadb.errors import InvalidDimensionException |
| from utils import * |
| from beschreibungen import * |
|
|
|
|
| |
| |
|
|
| from dotenv import load_dotenv, find_dotenv |
| _ = load_dotenv(find_dotenv()) |
|
|
|
|
|
|
| |
| |
| |
| |
| splittet = False |
|
|
| |
| |
| |
| |
| |
| |
| |
| |
|
|
|
|
|
|
| |
| HUGGINGFACEHUB_API_TOKEN = os.getenv("HF_ACCESS_READ") |
| OAI_API_KEY=os.getenv("OPENAI_API_KEY") |
| HEADERS = {"Authorization": f"Bearer {HUGGINGFACEHUB_API_TOKEN}"} |
| TAVILY_KEY = os.getenv("TAVILY_KEY") |
| os.environ["TAVILY_API_KEY"] = TAVILY_KEY |
|
|
|
|
| |
| |
| |
| |
| |
| MODEL_NAME= "gpt-4-1106-preview" |
| MODEL_NAME_IMAGE = "gpt-4-vision-preview" |
|
|
|
|
| |
| |
| |
| repo_id = "HuggingFaceH4/zephyr-7b-alpha" |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| MODEL_NAME_HF = "mistralai/Mixtral-8x7B-Instruct-v0.1" |
| MODEL_NAME_OAI_ZEICHNEN = "dall-e-3" |
| |
| API_URL = "https://api-inference.huggingface.co/models/stabilityai/stable-diffusion-2-1" |
|
|
| |
| |
| |
| os.environ["HUGGINGFACEHUB_API_TOKEN"] = HUGGINGFACEHUB_API_TOKEN |
|
|
|
|
| |
| |
| |
| |
| client = OpenAI() |
| general_assistant_file = client.beta.assistants.create(name="File Analysator",instructions=template, model="gpt-4-1106-preview",) |
| thread_file = client.beta.threads.create() |
| general_assistant_suche= openai_assistant_suche(client) |
|
|
| |
| |
| |
| |
| |
|
|
| |
| |
| def clear_all(): |
| return None, gr.Image(visible=False), [] |
|
|
| |
| |
| def add_text(chatbot, history, prompt, file): |
| if (file == None): |
| chatbot = chatbot +[(prompt, None)] |
| else: |
| if (prompt == ""): |
| chatbot=chatbot + [((file.name,), "Prompt fehlt!")] |
| else: |
| chatbot = chatbot +[((file.name,), None), (prompt, None)] |
|
|
| return chatbot, history, prompt, file, gr.Image(visible = False), "" |
|
|
| def add_text2(chatbot, prompt): |
| if (prompt == ""): |
| chatbot = chatbot + [("", "Prompt fehlt!")] |
| else: |
| chatbot = chatbot + [(prompt, None)] |
| print("chatbot nach add_text............") |
| print(chatbot) |
| return chatbot, prompt, "" |
| |
| |
| |
| def file_anzeigen(file): |
| ext = analyze_file(file) |
| if (ext == "png" or ext == "PNG" or ext == "jpg" or ext == "jpeg" or ext == "JPG" or ext == "JPEG"): |
| return gr.Image(width=47, visible=True, interactive = False, height=47, min_width=47, show_label=False, show_share_button=False, show_download_button=False, scale = 0.5), file, file |
| else: |
| return gr.Image(width=47, visible=True, interactive = False, height=47, min_width=47, show_label=False, show_share_button=False, show_download_button=False, scale = 0.5), "data/file.png", file |
| |
| def file_loeschen(): |
| return None, gr.Image(visible = False) |
|
|
| |
| |
| def cancel_outputing(): |
| reset_textbox() |
| return "Stop Done" |
|
|
| def reset_textbox(): |
| return gr.update(value=""),"" |
|
|
| |
| |
| |
| def umwandeln_fuer_anzeige(image): |
| buffer = io.BytesIO() |
| image.save(buffer, format='PNG') |
| return buffer.getvalue() |
|
|
|
|
|
|
| |
| |
| |
| def process_image(image_path, prompt): |
| |
| with open(image_path, "rb") as image_file: |
| encoded_string = base64.b64encode(image_file.read()).decode('utf-8') |
|
|
| |
| headers = { |
| "Content-Type": "application/json", |
| "Authorization": f"Bearer {OAI_API_KEY}" |
| } |
| payload = { |
| "model": MODEL_NAME_IMAGE, |
| "messages": [ |
| { |
| "role": "user", |
| "content": [ |
| { |
| "type": "text", |
| "text": llm_template + prompt |
| }, |
| { |
| "type": "image_url", |
| "image_url": { |
| "url": f"data:image/jpeg;base64,{encoded_string}" |
| } |
| } |
| ] |
| } |
| ], |
| "max_tokens": 300 |
| } |
| return headers, payload |
|
|
| |
| |
| def create_assistant_file(prompt, file): |
| global client, general_assistant_file |
| |
| file_neu = client.files.create(file=open(file,"rb",),purpose="assistants",) |
| |
| |
| |
| |
| updated_assistant = client.beta.assistants.update(general_assistant_file.id,tools=[{"type": "code_interpreter"}, {"type": "retrieval"}],file_ids=[file_neu.id],) |
| thread_file, run = create_thread_and_run(prompt, client, updated_assistant.id) |
| run = wait_on_run(run, thread_file, client) |
| response = get_response(thread_file, client, updated_assistant.id) |
| result = response.data[1].content[0].text.value |
| return result |
|
|
| |
| |
| def create_assistant_suche(prompt): |
| global client, general_assistant_suche |
|
|
| retriever = TavilySearchAPIRetriever(k=4) |
| result = retriever.invoke(prompt) |
|
|
| """ |
| #neues Thread mit akt. prompt dem Assistant hinzufügen |
| thread_suche, run = create_thread_and_run(prompt, client, general_assistant_suche.id) |
| run = wait_on_run(run, thread_suche, client) |
| response = get_response(thread_suche, client, general_assistant_suche.id) |
| result = response.data[1].content[0].text.value |
| """ |
| |
| |
| return result |
|
|
| |
| |
| |
| |
| |
| |
| def generate_auswahl(prompt, file, chatbot, history, rag_option, model_option, openai_api_key, k=3, top_p=0.6, temperature=0.5, max_new_tokens=4048, max_context_length_tokens=2048, repetition_penalty=1.3,): |
| global splittet |
| |
| |
| |
| if (rag_option == "An"): |
| |
| if not splittet: |
| splits = document_loading_splitting() |
| document_storage_chroma(splits) |
| db = document_retrieval_chroma2() |
| splittet = True |
| else: |
| db=None |
| splittet = False |
|
|
| |
| status = "Antwort der KI ..." |
| if (file == None): |
| result, status = generate_text(prompt, chatbot, history, rag_option, model_option, openai_api_key, db, k=3, top_p=0.6, temperature=0.5, max_new_tokens=4048, max_context_length_tokens=2048, repetition_penalty=1.3,) |
| history = history + [(prompt, result)] |
| else: |
| |
| |
| ext = analyze_file(file) |
| if (ext == "png" or ext == "PNG" or ext == "jpg" or ext == "jpeg" or ext == "JPG" or ext == "JPEG"): |
| result= generate_text_zu_bild(file, prompt, k, rag_option, chatbot, db) |
| else: |
| result = generate_text_zu_doc(file, prompt, k, rag_option, chatbot, db) |
| |
| history = history + [((file,), None),(prompt, result)] |
|
|
| chatbot[-1][1] = "" |
| for character in result: |
| chatbot[-1][1] += character |
| time.sleep(0.03) |
| yield chatbot, history, None, status |
| if shared_state.interrupted: |
| shared_state.recover() |
| try: |
| yield chatbot, history, None, "Stop: Success" |
| except: |
| pass |
| |
| |
| |
| def generate_bild(prompt, chatbot, model_option_zeichnen='HuggingFace', temperature=0.5, max_new_tokens=4048,top_p=0.6, repetition_penalty=1.3): |
| global client |
| if (model_option_zeichnen == "Stable Diffusion"): |
| print("Bild Erzeugung HF..............................") |
| |
| data = {"inputs": prompt} |
| response = requests.post(API_URL, headers=HEADERS, json=data) |
| print("fertig Bild") |
| result = response.content |
| |
| image = Image.open(io.BytesIO(result)) |
| image_64 = umwandeln_fuer_anzeige(image) |
| chatbot[-1][1]= "<img src='data:image/png;base64,{0}'/>".format(base64.b64encode(image_64).decode('utf-8')) |
| else: |
| print("Bild Erzeugung DallE..............................") |
| |
| response = client.images.generate(model="dall-e-3",prompt=prompt,size="1024x1024",quality="standard",n=1, response_format='b64_json') |
| |
| chatbot[-1][1] = "<img src='data:image/png;base64,{0}'/>".format(response.data[0].b64_json) |
| |
| return chatbot, "Antwort KI: Success" |
| |
|
|
| |
| |
| def generate_text_zu_bild(file, prompt, k, rag_option, chatbot, db): |
| global splittet |
| print("Text mit Bild ..............................") |
| print(file) |
| prompt_neu = prompt |
| if (rag_option == "An"): |
| print("Bild mit RAG..............................") |
| neu_text_mit_chunks = rag_chain2(prompt, db, k) |
| |
| |
| |
| prompt_neu = generate_prompt_with_history(neu_text_mit_chunks, chatbot) |
|
|
| headers, payload = process_image(file, prompt_neu) |
| response = requests.post("https://api.openai.com/v1/chat/completions", headers=headers, json=payload) |
| |
| data = response.json() |
| |
| result = data['choices'][0]['message']['content'] |
| return result |
|
|
|
|
| |
| |
| def generate_text_zu_doc(file, prompt, k, rag_option, chatbot, db): |
| global splittet |
| print("text mit doc ..............................") |
| |
| prompt_neu = prompt |
| if (rag_option == "An"): |
| print("Doc mit RAG..............................") |
| neu_text_mit_chunks = rag_chain2(prompt, db, k) |
| |
| |
| |
| prompt_neu = generate_prompt_with_history(neu_text_mit_chunks, chatbot) |
| |
| result = create_assistant_file(prompt_neu, file) |
| return result |
| |
|
|
| |
| |
| |
| def generate_text (prompt, chatbot, history, rag_option, model_option, openai_api_key, db, k=3, top_p=0.6, temperature=0.5, max_new_tokens=4048, max_context_length_tokens=2048, repetition_penalty=1.3,): |
| global splittet |
| suche_im_Netz="Antwort der KI ..." |
| print("Text pur..............................") |
| if (openai_api_key == "" or openai_api_key == "sk-"): |
| |
| |
| openai_api_key= OAI_API_KEY |
| if (rag_option is None): |
| raise gr.Error("Retrieval Augmented Generation ist erforderlich.") |
| if (prompt == ""): |
| raise gr.Error("Prompt ist erforderlich.") |
|
|
| |
| |
| |
| |
| |
| |
| |
| try: |
| |
| |
| |
| if (model_option == "OpenAI"): |
| |
| print("OpenAI Anfrage.......................") |
| llm = ChatOpenAI(model_name = MODEL_NAME, openai_api_key = openai_api_key, temperature=temperature) |
| |
| if (rag_option == "An"): |
| history_text_und_prompt = generate_prompt_with_history(prompt, chatbot) |
| else: |
| history_text_und_prompt = generate_prompt_with_history_openai(prompt, chatbot) |
| else: |
| |
| print("HF Anfrage.......................") |
| llm = HuggingFaceHub(repo_id=repo_id, model_kwargs={"temperature": 0.5, "max_length": 128}) |
| |
| |
| |
| print("HF") |
| |
| history_text_und_prompt = generate_prompt_with_history(prompt, chatbot) |
| |
| |
| if (rag_option == "An"): |
| print("LLM aufrufen mit RAG: ...........") |
| result = rag_chain(llm, history_text_und_prompt, db) |
| |
| |
| |
| |
| |
| else: |
| splittet = False |
| print("LLM aufrufen ohne RAG: ...........") |
| result = create_assistant_suche(history_text_und_prompt) |
|
|
| |
| if is_response_similar(result): |
| print("Suche im Netz: ...........") |
| suche_im_Netz="Antwort aus dem Internet ..." |
| result = create_assistant_suche(history_text_und_prompt) |
| |
| except Exception as e: |
| raise gr.Error(e) |
|
|
| return result, suche_im_Netz |
|
|
| |
| |
| |
| |
| |
| |
| |
|
|
| |
| |
|
|
| def vote(data: gr.LikeData): |
| if data.liked: print("You upvoted this response: " + data.value) |
| else: print("You downvoted this response: " + data.value) |
|
|
|
|
| print ("Start GUIneu") |
| with open("custom.css", "r", encoding="utf-8") as f: |
| customCSS = f.read() |
|
|
| |
| additional_inputs = [ |
| gr.Slider(label="Temperature", value=0.65, minimum=0.0, maximum=1.0, step=0.05, interactive=True, info="Höhere Werte erzeugen diversere Antworten", visible=True), |
| gr.Slider(label="Max new tokens", value=1024, minimum=0, maximum=4096, step=64, interactive=True, info="Maximale Anzahl neuer Tokens", visible=True), |
| gr.Slider(label="Top-p (nucleus sampling)", value=0.6, minimum=0.0, maximum=1, step=0.05, interactive=True, info="Höhere Werte verwenden auch Tokens mit niedrigerer Wahrscheinlichkeit.", visible=True), |
| gr.Slider(label="Repetition penalty", value=1.2, minimum=1.0, maximum=2.0, step=0.05, interactive=True, info="Strafe für wiederholte Tokens", visible=True) |
| ] |
| with gr.Blocks(css=customCSS, theme=small_and_beautiful_theme) as demo: |
| |
| |
| history = gr.State([]) |
| |
| user_question = gr.State("") |
| |
| user_question2 = gr.State("") |
| attached_file = gr.State(None) |
| status_display = gr.State("") |
| status_display2 = gr.State("") |
| |
| |
| |
| gr.Markdown(description_top) |
| with gr.Tab("LI Chatbot"): |
| with gr.Row(): |
| |
| status_display = gr.Markdown("Antwort der KI ...", visible = False, elem_id="status_display") |
| with gr.Row(): |
| with gr.Column(scale=5): |
| with gr.Row(): |
| chatbot = gr.Chatbot(elem_id="li-chat",show_copy_button=True) |
| with gr.Row(): |
| with gr.Column(scale=12): |
| user_input = gr.Textbox( |
| show_label=False, placeholder="Gib hier deinen Prompt ein...", |
| container=False |
| ) |
| with gr.Column(min_width=70, scale=1): |
| submitBtn = gr.Button("Senden") |
| with gr.Column(min_width=70, scale=1): |
| cancelBtn = gr.Button("Stop") |
| with gr.Row(): |
| |
| image_display = gr.Image( visible=False) |
| upload = gr.UploadButton("📁", file_types=["image", "pdf", "docx", "pptx", "xlsx"], scale = 10) |
| emptyBtn = gr.ClearButton([user_input, chatbot, history, attached_file, image_display], value="🧹 Neue Session", scale=10) |
| |
| with gr.Column(): |
| with gr.Column(min_width=50, scale=1): |
| with gr.Tab(label="Parameter Einstellung"): |
| |
| rag_option = gr.Radio(["Aus", "An"], label="LI Erweiterungen (RAG)", value = "Aus") |
| model_option = gr.Radio(["OpenAI", "HuggingFace"], label="Modellauswahl", value = "OpenAI") |
|
|
| |
| top_p = gr.Slider( |
| minimum=-0, |
| maximum=1.0, |
| value=0.95, |
| step=0.05, |
| interactive=True, |
| label="Top-p", |
| visible=False, |
| ) |
| temperature = gr.Slider( |
| minimum=0.1, |
| maximum=2.0, |
| value=0.5, |
| step=0.1, |
| interactive=True, |
| label="Temperature", |
| visible=False |
| ) |
| max_length_tokens = gr.Slider( |
| minimum=0, |
| maximum=512, |
| value=512, |
| step=8, |
| interactive=True, |
| label="Max Generation Tokens", |
| visible=False, |
| ) |
| max_context_length_tokens = gr.Slider( |
| minimum=0, |
| maximum=4096, |
| value=2048, |
| step=128, |
| interactive=True, |
| label="Max History Tokens", |
| visible=False, |
| ) |
| repetition_penalty=gr.Slider(label="Repetition penalty", value=1.2, minimum=1.0, maximum=2.0, step=0.05, interactive=True, info="Strafe für wiederholte Tokens", visible=False) |
| anzahl_docs = gr.Slider(label="Anzahl Dokumente", value=3, minimum=1, maximum=10, step=1, interactive=True, info="wie viele Dokumententeile aus dem Vektorstore an den prompt gehängt werden", visible=False) |
| openai_key = gr.Textbox(label = "OpenAI API Key", value = "sk-", lines = 1, visible = False) |
| |
| |
| |
| with gr.Tab("LI Zeichnen"): |
| with gr.Row(): |
| |
| status_display2 = gr.Markdown("Success", visible = False, elem_id="status_display") |
| |
| with gr.Row(): |
| with gr.Column(scale=5): |
| with gr.Row(): |
| chatbot_bild = gr.Chatbot(elem_id="li-zeichnen",show_copy_button=True, show_share_button=True) |
| with gr.Row(): |
| with gr.Column(scale=12): |
| user_input2 = gr.Textbox( |
| show_label=False, placeholder="Gib hier deinen Prompt ein...", |
| container=False |
| ) |
| with gr.Column(min_width=70, scale=1): |
| submitBtn2 = gr.Button("Senden") |
| |
| |
| with gr.Row(): |
| emptyBtn2 = gr.ClearButton([user_input, chatbot_bild], value="🧹 Neue Session", scale=10) |
| |
| with gr.Column(): |
| with gr.Column(min_width=50, scale=1): |
| with gr.Tab(label="Parameter Einstellung"): |
| |
| model_option_zeichnen = gr.Radio(["Stable Diffusion","DallE"], label="Modellauswahl", value = "Stable Diffusion") |
|
|
|
|
| gr.Markdown(description) |
| |
| |
| |
| |
| |
| |
| |
| predict_args = dict( |
| fn=generate_auswahl, |
| inputs=[ |
| user_question, |
| attached_file, |
| chatbot, |
| history, |
| rag_option, |
| model_option, |
| openai_key, |
| anzahl_docs, |
| top_p, |
| temperature, |
| max_length_tokens, |
| max_context_length_tokens, |
| repetition_penalty |
| ], |
| outputs=[chatbot, history, attached_file, status_display], |
| show_progress=True, |
| ) |
| |
| reset_args = dict( |
| fn=reset_textbox, inputs=[], outputs=[user_input, status_display] |
| ) |
|
|
| |
| transfer_input_args = dict( |
| fn=add_text, inputs=[chatbot, history, user_input, attached_file], outputs=[chatbot, history, user_question, attached_file, image_display , user_input], show_progress=True |
| ) |
|
|
| predict_event1 = user_input.submit(**transfer_input_args, queue=False,).then(**predict_args) |
| predict_event2 = submitBtn.click(**transfer_input_args, queue=False,).then(**predict_args) |
| predict_event3 = upload.upload(file_anzeigen, [upload], [image_display, image_display, attached_file] ) |
| emptyBtn.click(clear_all, [], [attached_file, image_display, history]) |
| image_display.select(file_loeschen, [], [attached_file, image_display]) |
| |
| |
| cancelBtn.click(cancel_outputing, [], [status_display], cancels=[predict_event1,predict_event2, predict_event3]) |
|
|
| |
| |
| predict_args2 = dict( |
| fn=generate_bild, |
| inputs=[ |
| user_question2, |
| chatbot_bild, |
| model_option_zeichnen, |
| |
| ], |
| outputs=[chatbot_bild, status_display2], |
| show_progress=True, |
| ) |
| transfer_input_args2 = dict( |
| fn=add_text2, inputs=[chatbot_bild, user_input2], outputs=[chatbot_bild, user_question2, user_input2], show_progress=True |
| ) |
| predict_event2_1 = user_input2.submit(**transfer_input_args2, queue=False,).then(**predict_args2) |
| predict_event2_2 = submitBtn2.click(**transfer_input_args2, queue=False,).then(**predict_args2) |
| |
| |
| |
| |
| |
| |
| |
| demo.title = "LI-ChatBot" |
| demo.queue().launch(debug=True) |
|
|
|
|
|
|
|
|
|
|
|
|