import torch
import gradio as gr
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig

model_id = "ALLaM-AI/ALLaM-7B-Instruct-preview"

bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_compute_dtype=torch.float16,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_use_double_quant=True
)

tokenizer = AutoTokenizer.from_pretrained(model_id)
tokenizer.pad_token = tokenizer.eos_token

model = AutoModelForCausalLM.from_pretrained(
    model_id,
    quantization_config=bnb_config,
    device_map="auto"
)

def generate_answer(question):
    prompt = f"أجب بالعربية الفصحى فقط:\n{question}\n"

    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

    output = model.generate(
        **inputs,
        max_new_tokens=200,
        temperature=0.7,
        do_sample=True,
        pad_token_id=tokenizer.eos_token_id
    )

    return tokenizer.decode(output[0], skip_special_tokens=True)

demo = gr.Interface(
    fn=generate_answer,
    inputs="text",
    outputs="text",
)

demo.launch()