import gradio as gr
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer
from huggingface_hub import snapshot_download

# Set device to CPU
device = "cpu"
repo_id = 'amgadhasan/phi-2'
model_path = snapshot_download(repo_id=repo_id, repo_type="model", local_dir="./phi-2", use_auth_token=False)

tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)

# Set default dtype to float32 for compatibility with CPU
torch.set_default_dtype(torch.float32)

model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto", trust_remote_code=True)

def generate(prompt):
    inputs = tokenizer(prompt, return_tensors="pt").to(device)
    outputs = model.generate(**inputs, max_length=200)
    completion = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return completion

def ask_question(user_question):
    if user_question.lower() == 'quit':
        return "Session ended. Goodbye!"
    else:
        # Here, we're explicitly setting the context for an academic answer.
        prompt = f"Academic response to the question about basic science subjects: {user_question}"
        answer = generate(prompt)
        return answer

iface = gr.Interface(fn=ask_question, inputs="text", outputs="text")
iface.launch(share=True)