app.py

from pprint import pprint, pformat
import os

import gradio as gr
import click
from rasa.nlu.model import Interpreter

from transformers import AutoTokenizer, AutoModelForSequenceClassification


# Rasa intent + entity extractor
RASA_MODEL_PATH = "woz_nlu_agent/models/nlu"
interpreter = None

# OOS classifier
tokenizer = AutoTokenizer.from_pretrained("msamogh/autonlp-cai-out-of-scope-649919116")
model = AutoModelForSequenceClassification.from_pretrained("msamogh/autonlp-cai-out-of-scope-649919116")


MODEL_TYPES = {
    "Out-of-scope classifier": "oos",
    "Intent classifier": "intent_transformer",
    "Intent and Entity extractor": "rasa_intent_entity"
}

def predict(model_type, input):
    if MODEL_TYPES[model_type] == "rasa_intent_entity":
        return rasa_predict(input)
    elif MODEL_TYPES[model_type] == "oos":
        return oos_predict(input)
    elif MODEL_TYPES[model_type] == "intent_transformer":
        return "TODO:: intent_transformer"

    
def oos_predict(input):
    inputs = tokenizer(input, return_tensors="pt")
    outputs = model(**inputs).logits
    outputs = torch.softmax(torch.tensor(outputs), dim=-1)[0]
    return str({"In scope": outputs[1], "Out of scope": outputs[0]})
    

def rasa_predict(input):

    def rasa_output(text):
        message = str(text).strip()
        result = interpreter.parse(message)
        return result

    response = rasa_output(input)
    
    del response["response_selector"]
    response["intent_ranking"] = response["intent_ranking"][:3]
    if "id" in response["intent"]:
        del response["intent"]["id"]
    for i in response["intent_ranking"]:
        if "id" in i:
            del i["id"]
    for e in response["entities"]:
        if "extractor" in e:
            del e["extractor"]
        if "start" in e and "end" in e:
            del e["start"]
            del e["end"]

    return pformat(response, indent=4)


def main():
    global interpreter
    print("Loading model...")
    print(os.listdir("woz_nlu_agent/models/nlu"))
    print(open("woz_nlu_agent/models/nlu/metadata.json", "r").read())
    import json
    print(json.load(open("woz_nlu_agent/models/nlu/metadata.json", "r")))
    
    interpreter = Interpreter.load(RASA_MODEL_PATH)
    print("Model loaded.")
    iface = gr.Interface(fn=predict, inputs=[gr.inputs.Dropdown(list(MODEL_TYPES.keys())), "text"], outputs="text")
    iface.launch()


if __name__ == "__main__":
    main()