Spaces:
Running
Running
| ''' | |
| This script calls the model from openai api to predict the next few words. | |
| ''' | |
| import os | |
| # os.system("pip install --upgrade pip") | |
| from pprint import pprint | |
| # os.system("pip install git+https://github.com/openai/whisper.git") | |
| import sys | |
| # print("Sys: ", sys.executable) | |
| # os.system("pip install openai") | |
| import openai | |
| import gradio as gr | |
| import whisper | |
| from transformers import pipeline | |
| import torch | |
| from transformers import AutoModelForCausalLM | |
| from transformers import AutoTokenizer | |
| import time | |
| EXAMPLE_PROMPT = """This is a tool for helping someone with memory issues remember the next word. | |
| The predictions follow a few rules: | |
| 1) The predictions are suggestions of ways to continue the transcript as if someone forgot what the next word was. | |
| 2) The predictions do not repeat themselves. | |
| 3) The predictions focus on suggesting nouns, adjectives, and verbs. | |
| 4) The predictions are related to the context in the transcript. | |
| EXAMPLES: | |
| Transcript: Tomorrow night we're going out to | |
| Prediction: The Movies, A Restaurant, A Baseball Game, The Theater, A Party for a friend | |
| Transcript: I would like to order a cheeseburger with a side of | |
| Prediction: Frnech fries, Milkshake, Apple slices, Side salad, Extra katsup | |
| Transcript: My friend Savanah is | |
| Prediction: An elecrical engineer, A marine biologist, A classical musician | |
| Transcript: I need to buy a birthday | |
| Prediction: Present, Gift, Cake, Card | |
| Transcript: """ | |
| # whisper model specification | |
| model = whisper.load_model("tiny") | |
| # openai.api_key = os.environ["Openai_APIkey"] | |
| def debug_inference(audio, prompt, model, temperature, state=""): | |
| breakpoint() | |
| # load audio data | |
| audio = whisper.load_audio(audio) | |
| # ensure sample is in correct format for inference | |
| audio = whisper.pad_or_trim(audio) | |
| # generate a log-mel spetrogram of the audio data | |
| mel = whisper.log_mel_spectrogram(audio) | |
| _, probs = model.detect_language(mel) | |
| # decode audio data | |
| options = whisper.DecodingOptions(fp16 = False) | |
| # transcribe speech to text | |
| result = whisper.decode(model, mel, options) | |
| print("result pre gp model from whisper: ", result, ".text ", result.text, "and the data type: ", type(result.text)) | |
| text = prompt + result.text + "\nPrediction: " | |
| response = openai.Completion.create( | |
| model=model, | |
| prompt=text, | |
| temperature=temperature, | |
| max_tokens=8, | |
| n=5) | |
| infers = [] | |
| temp = [] | |
| infered=[] | |
| for i in range(5): | |
| print("print1 ", response['choices'][i]['text']) | |
| temp.append(response['choices'][i]['text']) | |
| print("print2: infers ", infers) | |
| print("print3: Responses ", response) | |
| print("Object type of response: ", type(response)) | |
| #infered = list(map(lambda x: x.split(',')[0], infers)) | |
| #print("Infered type is: ", type(infered)) | |
| infers = list(map(lambda x: x.replace("\n", ""), temp)) | |
| #infered = list(map(lambda x: x.split(','), infers)) | |
| return result.text, state, infers, text | |
| # get audio from microphone | |
| gr.Interface( | |
| fn=debug_inference, | |
| inputs=[gr.inputs.Audio(source="microphone", type="filepath"), | |
| gr.inputs.Textbox(lines=15, placeholder="Enter a prompt here"), | |
| gr.inputs.Dropdown(["text-ada-001", "text-davinci-002", "text-davinci-003", "gpt-3.5-turbo"], label="Model"), | |
| gr.inputs.Slider(minimum=0.0, maximum=1.0, default=0.8, step=0.1, label="Temperature"), | |
| "state" | |
| ], | |
| outputs=["textbox","state","textbox", "textbox"], | |
| examples=[["example_in-the-mood-to-eat.m4a", EXAMPLE_PROMPT, "text-ada-001", 0.8, ""],["","","",0.9,""]], | |
| live=False).launch() |