Spaces:
Sleeping
Sleeping
from __future__ import annotations | |
import gradio as gr | |
import torch | |
import os | |
import polars as pl | |
import re | |
import json | |
from datetime import datetime, timezone, timedelta | |
from optimum.pipelines import pipeline | |
from optimum.onnxruntime import ORTModelForSequenceClassification | |
from transformers import AutoTokenizer | |
from hf_dataset_saver import HuggingFaceDatasetSaver | |
# Get environment variable | |
hf_token = os.getenv('HF_TOKEN') | |
if torch.cuda.is_available(): | |
print("GPU is enabled.") | |
print("device count: {}, current device: {}".format(torch.cuda.device_count(), torch.cuda.current_device())) | |
else: | |
print("GPU is not enabled.") | |
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") | |
# Prepare logger for flagging | |
hf_writer = HuggingFaceDatasetSaver(hf_token, "crowdsourced-sentiment_analysis") | |
# Prepare model | |
tokenizer = AutoTokenizer.from_pretrained("xlm-roberta-base", token=hf_token) | |
model = ORTModelForSequenceClassification.from_pretrained("arcleife/roberta-sentiment-id-onnx", num_labels=3, token=hf_token).to(device) | |
pipe = pipeline("text-classification", model=model, tokenizer=tokenizer, device=device, return_token_type_ids=False, accelerator="ort") | |
def get_label(result): | |
if result[0]['label'] == "LABEL_0": | |
return "POSITIVE" | |
elif result[0]['label'] == "LABEL_1": | |
return "NEUTRAL" | |
else: | |
return "NEGATIVE" | |
def text_classification(text): | |
result = pipe(text) | |
sentiment_label = get_label(result) | |
sentiment_score = result[0]['score'] | |
return sentiment_label, sentiment_score | |
examples=["Makanannya ga enak ini", "Nyaman ya tempatnya"] | |
io = gr.Interface(fn=text_classification, | |
inputs=gr.Textbox(lines=2, label="Text", placeholder="Enter text here..."), | |
outputs=["text", "number"], | |
title="Text Classification", | |
description="Enter a text and see the text classification result!", | |
examples=examples, | |
# flagging_mode="manual", | |
# flagging_options=["TOXIC", "NONTOXIC"], | |
# flagging_callback=hf_writer | |
) | |
io.launch(inline=False) |