app.py

import gradio as gr
import spaces
from transformers import pipeline
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

model_name = "sapienzanlp/Minerva-7B-instruct-v1.0"
model = AutoModelForCausalLM.from_pretrained(
    model_name,
    torch_dtype=torch.bfloat16,
    device_map="auto"
)
tokenizer = AutoTokenizer.from_pretrained(model_name)

classifier = pipeline("text-classification", model="saiteki-kai/QA-DeBERTa-v3-large")

@spaces.GPU()
def generate(prompts: list[str]) -> tuple[list[str], list[dict[str, float]]]:
    messages = [[{"role": "user", "content": message}] for message in prompts]

    texts = tokenizer.apply_chat_template(
        messages,
        tokenize=False,
        add_generation_prompt=True
    )
    model_inputs = tokenizer(texts, padding=True, return_tensors="pt").to(model.device)
    generated_ids = model.generate(
        **model_inputs,
        do_sample=False,
        temperature=0,
        repetition_penalty=1.0,
        max_new_tokens=512,
    )
    generated_ids = [
        output_ids[len(input_ids):] for input_ids, output_ids in zip(model_inputs.input_ids, generated_ids)
    ]
    responses = tokenizer.batch_decode(generated_ids, skip_special_tokens=True)

    return responses, classifier([text + "[SEP]" + response for text, response in zip(texts, responses)])

with gr.Blocks() as demo:
    gr.Markdown("Welcome")
    gr.api(generate, api_name="predict", batch=True)

demo.launch()