Armanul's picture
Update app.py
1e6b6c9 verified
import gradio as gr
import onnxruntime as rt
from transformers import AutoTokenizer
import torch
import json
tokenizer = AutoTokenizer.from_pretrained("distilroberta-base")
with open("dataset_types_encoded.json", "r") as fp:
encode_category_types = json.load(fp)
categories = list(encode_category_types.keys())
inf_session = rt.InferenceSession('dataset-classifier-distilroberta-quantized.onnx')
input_name = inf_session.get_inputs()[0].name
output_name = inf_session.get_outputs()[0].name
def classify_dataset_type(description):
input_ids = tokenizer(description)['input_ids'][:512]
logits = inf_session.run([output_name], {input_name: [input_ids]})[0]
logits = torch.FloatTensor(logits)
probs = torch.sigmoid(logits)[0]
return dict(zip(categories, map(float, probs)))
label = gr.outputs.Label(num_top_classes=3)
iface = gr.Interface(fn=classify_dataset_type, inputs="text", outputs=label)
# iface = gr.Interface(fn=classify_dataset_type, inputs="textbox", outputs=gr.Label())
iface.launch(inline=False, share=True)