app.py

import gradio as gr
import torch
import torch.nn.functional as F
from sentence_transformers import SentenceTransformer

# Load the model
revision = None  # Replace with the specific revision to ensure reproducibility if the model is updated.
model = SentenceTransformer("avsolatorio/GIST-small-Embedding-v0", revision=revision)

# Precompute reference embeddings
ref_texts = [
    "Theatro App: Hello John. Hey John. Hi John. Call John",
    "Theatro App: Message John. Message for John. Leave a message for John",
    "Theatro App: Play messages. Listen to messages",
    "Theatro App: What time is it?",
    "Theatro App: What time is it?",
    "Theatro App: Cashier Backup. Backup Cashier. Register backup. Register assistance.",
    "Theatro App: repeat",
    "Theatro App: Check inventory",
    "Theatro App: Check Sales",
    "Theatro App: Curbside Pickup",
    "Theatro App: Replay last message.",
    "Theatro App: Post it. Post it for group"
    "Theatro App: Announcement. Announcement for the group",
    "Open question: This is about products sold in TractorSupply.",
    "Open question: This is about pet care.",
    "Open question: What is the weather like?",
    "Open question: What's 15% off from $79.99?",
    "Open question: Can you look up the skew for 1091784?",
]

ref_embeddings = model.encode(ref_texts, convert_to_tensor=True)

def find_query_type(query):
    query_embeddings = model.encode([query], convert_to_tensor=True)
    scores = F.cosine_similarity(query_embeddings, ref_embeddings, dim=-1)
    max_index = torch.argmax(scores).item()
    ref_text = ref_texts[max_index]
    query_type = ref_text.split(": ")[0]
    return query_type

import gradio as gr
def predict(query):
    query_type = find_query_type(query)
    return query_type

iface = gr.Interface(fn=predict,
                     inputs=gr.Textbox(lines=2, placeholder="Enter your query here..."),
                     outputs="text",
                     title="Query Type Classifier",
                     description="This model classifies the type of your query. Just input your query and get the predicted category.")

iface.launch()