Spaces:
Running
Running
| #Praise Jesus | |
| #Stable version working with Llama but not satisfied with poor output | |
| import gradio as gr | |
| from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline | |
| import torch | |
| import docx2txt | |
| import os | |
| # Authenticate with Hugging Face using an environment variable token if set | |
| hf_token = os.getenv('HF_TOKEN') | |
| if hf_token: | |
| from huggingface_hub import login | |
| login(hf_token) | |
| # Whisper model for audio transcription | |
| whisper_model = pipeline("automatic-speech-recognition", model="openai/whisper-large") | |
| # LLaMA 3.2 model for text processing | |
| llama_model_id = "meta-llama/Llama-3.2-3B-Instruct" | |
| tokenizer = AutoTokenizer.from_pretrained(llama_model_id, token=hf_token) | |
| llama_model = AutoModelForCausalLM.from_pretrained(llama_model_id, torch_dtype=torch.bfloat16, token=hf_token) | |
| # Function to generate response using LLaMA 3.2 model | |
| def get_llama_response(input_text): | |
| # Ensure input is detailed enough | |
| if len(input_text.split()) < 10: | |
| return "Please provide a more detailed user story to help generate relevant needs and wants." | |
| # Define prompt for LLaMA model | |
| prompt = f""" | |
| Based on the user story "{input_text}", extract any unarticulated needs and wants. | |
| Only provide essential needs and wants directly relevant to the given story. | |
| Do not speculate or over-extrapolate. | |
| """ | |
| # Process the prompt with LLaMA 3.2 | |
| inputs = tokenizer(prompt, return_tensors="pt") | |
| llama_output = llama_model.generate(**inputs, max_new_tokens=100) | |
| response_text = tokenizer.decode(llama_output[0], skip_special_tokens=True) | |
| return response_text | |
| # Main processing function for Gradio interface | |
| def process_input(user_story=None, user_audio=None, user_file=None): | |
| # Process audio input if provided | |
| if user_audio: | |
| transcription = whisper_model(user_audio)["text"] | |
| user_story = transcription | |
| # Process file input if provided and if text is empty | |
| if user_file and not user_story: | |
| user_story = docx2txt.process(user_file) | |
| # Ensure there's text to process | |
| if not user_story: | |
| return "Please provide a user story, an audio file, or upload a Word file." | |
| # Generate response with LLaMA 3.2 | |
| llama_response = get_llama_response(user_story) | |
| return f"LLaMA Output:\n{llama_response}" | |
| # Gradio interface with text, audio, and file inputs | |
| interface = gr.Interface( | |
| fn=process_input, | |
| inputs=[ | |
| gr.Textbox(label="User Story (Text Input)", placeholder="Enter your user story here..."), | |
| gr.Audio(type="filepath", label="User Story (Audio Input)"), | |
| gr.File(label="Upload Word File (.docx)") # Removed `optional=True` | |
| ], | |
| outputs="text", | |
| title="Multimodal Needs & Wants Extractor", | |
| description="**Author:** VictorDaniel\n\nEnter a detailed user story or upload an audio/Word file to extract the unarticulated needs and wants.", | |
| examples=[ | |
| ["The user often speaks about wanting to improve their health but is hesitant to join a gym."] | |
| ] | |
| ) | |
| # Launch the Gradio app | |
| interface.launch() | |
| #app2 |