app.py

import pandas as pd
import numpy as np
import gradio as gr
import torch
from transformers import AutoModelForMultipleChoice, AutoTokenizer
from huggingface_hub import hf_hub_url, Repository
model_id="/kaggle/input/deberta-v3-large-hf-weights"
# Load the model and tokenizer

model = AutoModelForMultipleChoice.from_pretrained(model_id)
tokenizer = AutoTokenizer.from_pretrained(model_id)

# Define the preprocessing function
def preprocess(sample):
    first_sentences = [sample["prompt"]] * 5
    second_sentences = [sample[option] for option in "ABCDE"]
    tokenized_sentences = tokenizer(first_sentences, second_sentences, truncation=True, padding=True, return_tensors="pt")
    sample["input_ids"] = tokenized_sentences["input_ids"]
    sample["attention_mask"] = tokenized_sentences["attention_mask"]
    return sample

# Define the prediction function
def predict(data):
    inputs = torch.stack(data["input_ids"])
    masks = torch.stack(data["attention_mask"])
    with torch.no_grad():
        logits = model(inputs, attention_mask=masks).logits
    predictions_as_ids = torch.argsort(-logits, dim=1)
    answers = np.array(list("ABCDE"))[predictions_as_ids.tolist()]
    return ["".join(i) for i in answers[:, :3]]

# Create the Gradio interface
iface = gr.Interface(
    fn=predict,
    inputs=gr.Interface.DataType.json,
    outputs=gr.outputs.Label(num_top_classes=3),
    live=True,
    examples=[
        {"prompt": "This is the prompt", "A": "Option A text", "B": "Option B text", "C": "Option C text", "D": "Option D text", "E": "Option E text"}
    ],
    title="LLM Science Exam Demo",
    description="Enter the prompt and options (A to E) below and get predictions.",
)

# Run the interface
iface.launch()
import pandas as pd
import numpy as np
import gradio as gr
import torch
from transformers import AutoModelForMultipleChoice, AutoTokenizer
from huggingface_hub import hf_hub_url, Repository

# Load the model and tokenizer
model_path = "my_model"
model = AutoModelForMultipleChoice.from_pretrained(model_path)
tokenizer = AutoTokenizer.from_pretrained(model_path)

# Define the preprocessing function
def preprocess(sample):
    first_sentences = [sample["prompt"]] * 5
    second_sentences = [sample[option] for option in "ABCDE"]
    tokenized_sentences = tokenizer(first_sentences, second_sentences, truncation=True, padding=True, return_tensors="pt")
    sample["input_ids"] = tokenized_sentences["input_ids"]
    sample["attention_mask"] = tokenized_sentences["attention_mask"]
    return sample

# Define the prediction function
def predict(data):
    inputs = torch.stack(data["input_ids"])
    masks = torch.stack(data["attention_mask"])
    with torch.no_grad():
        logits = model(inputs, attention_mask=masks).logits
    predictions_as_ids = torch.argsort(-logits, dim=1)
    answers = np.array(list("ABCDE"))[predictions_as_ids.tolist()]
    return ["".join(i) for i in answers[:, :3]]

# Create the Gradio interface
iface = gr.Interface(
    fn=predict,
    inputs=gr.Interface.DataType.json,
    outputs=gr.outputs.Label(num_top_classes=3),
    live=True,
    examples=[
        {"prompt": "This is the prompt", "A": "Option A text", "B": "Option B text", "C": "Option C text", "D": "Option D text", "E": "Option E text"}
    ],
    title="LLM Science Exam Demo",
    description="Enter the prompt and options (A to E) below and get predictions.",
)

# Run the interface
iface.launch(share=True)