app.py

import streamlit as st
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline
from peft import PeftModel
import torch

@st.cache_resource
def load_model():
    base_model = "deepseek-ai/deepseek-coder-5.7b-base"
    adapter_path = "faizabenatmane/deepseek-coder-5.7bmqa-finetuned"

    tokenizer = AutoTokenizer.from_pretrained(base_model)
    base = AutoModelForCausalLM.from_pretrained(base_model, torch_dtype=torch.float16, device_map="auto")

    model = PeftModel.from_pretrained(base, adapter_path)
    model = model.merge_and_unload()

    pipe = pipeline("text-generation", model=model, tokenizer=tokenizer)
    return pipe

generator = load_model()

st.title("🧠 DeepSeek QA (Generation)")
text = st.text_area("Ask a coding or general question:", height=200)

if st.button("Generate Answer"):
    with st.spinner("Generating..."):
        prompt = f"Question: {text}\nAnswer:"
        output = generator(prompt, max_new_tokens=100, do_sample=False)[0]["generated_text"]
        answer = output.split("Answer:")[-1].strip()

        st.subheader("Generated Answer")
        st.success(answer)