import transformers
from transformers import AutoConfig, AutoTokenizer, AutoModelForSequenceClassification
import torch
import gradio as gr

model_dir = "./experiments/checkpoint-382"

config = AutoConfig.from_pretrained(model_dir, num_labels=3, finetuning_task="text-classification")
tokenizer = AutoTokenizer.from_pretrained(model_dir)
model = AutoModelForSequenceClassification.from_pretrained(model_dir, config=config)

def inference(input_text):
    inputs = tokenizer.batch_encode_plus([input_text], return_tensors="pt", max_length=512, truncation=True, padding="max_length")

    with torch.no_grad():
        logits = model(**inputs)["logits"]

    predicted_class = torch.argmax(logits, dim=1).item()
    output = model.config.id2label[predicted_class]
    return output

with gr.Blocks(css=""".message.svelte-w6rprc.svelte-w6rprc.svelte-w6rprc {font-size: 1.5em; padding: 1em; text-align: center;} #component-21 > div.wrap.svelte-w6rprc {height: 600px;}""") as demo:
    with gr.Row():
        with gr.Column():
            input_text = gr.Textbox(label="Input Text", scale=2, container=False)
            answer = gr.Output(label="Output", lines=0)
            generate_btn = gr.Button(text="Generate", type="primary", scale=2)

    inputs = [input_text]
    outputs = [answer]

    generate_btn.click(fn=inference, inputs=inputs, outputs=outputs, show_progress=True)

    examples = [
        ["I love this movie!"],
        ["I hate this movie!"],
        ["I feel neutral about this movie!"]
    ]

demo.queue()
demo.launch()