File size: 2,614 Bytes
0a476ff
 
 
 
 
 
 
 
 
 
19e70cb
cc814c9
0a476ff
19e70cb
cc814c9
0a476ff
19e70cb
 
0a476ff
 
 
 
 
19e70cb
 
0a476ff
19e70cb
 
0a476ff
 
cc814c9
93894a3
cc814c9
0a476ff
cc814c9
 
 
 
 
 
 
93894a3
cc814c9
 
 
 
 
 
 
93894a3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
cc814c9
 
0a476ff
cc814c9
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
import streamlit as st
import torch
from transformers import BartTokenizer, BartForConditionalGeneration
from peft import PeftModel
import textstat

@st.cache_resource
def load_model():
    base = BartForConditionalGeneration.from_pretrained(
        "facebook/bart-large-cnn",
        torch_dtype=torch.float32,
        device_map=None
    )
    
    model = PeftModel.from_pretrained(base, "./checkpoint")
    tokenizer = BartTokenizer.from_pretrained("facebook/bart-large-cnn")
    
    model.to("cpu")
    model.eval()
    return tokenizer, model

def simplify(text, tokenizer, model):
    prompt = f"simplify: {text}"
    inputs = tokenizer(prompt, return_tensors="pt", max_length=512, truncation=True)
    
    with torch.inference_mode():
        outputs = model.generate(**inputs, max_new_tokens=256, num_beams=4, early_stopping=True)
    
    return tokenizer.decode(outputs[0], skip_special_tokens=True)

st.set_page_config(page_title="Legaleaze", layout="wide")
st.title("Legaleaze: Legal Text Simplifier")
st.caption("BART-Large + LoRA | 121k steps on asylum cases")

try:
    tokenizer, model = load_model()
    
    col1, col2 = st.columns(2)
    
    with col1:
        st.subheader("Complex Legal Text")
        text = st.text_area("", height=300, placeholder="Paste legal text here...", key="input")
        btn = st.button("Simplify", type="primary", use_container_width=True)
    
    with col2:
        st.subheader("Simplified Output")
        if btn and text.strip():
            with st.spinner("Simplifying (30s on CPU)..."):
                result = simplify(text, tokenizer, model)
                st.session_state['result'] = result
                st.session_state['original_text'] = text
        
        if 'result' in st.session_state:
            # Editable output
            simplified = st.text_area("", value=st.session_state['result'], height=300, key="output")
            
            # Copy button
            if st.button("📋 Copy to Clipboard", use_container_width=True):
                st.write("Copy the text above manually (browser limitation)")
            
            st.divider()
            m1, m2, m3 = st.columns(3)
            orig = textstat.flesch_kincaid_grade(st.session_state['original_text'])
            simp = textstat.flesch_kincaid_grade(simplified)
            m1.metric("Original FKGL", f"{orig:.1f}")
            m2.metric("Simplified FKGL", f"{simp:.1f}")
            m3.metric("Improvement", f"{((orig-simp)/orig*100):.0f}%")
        else:
            st.info("Simplified text appears here")

except Exception as e:
    st.error(f"Error: {e}")