Update app.py

app.py

@@ -5,16 +5,15 @@ from peft import PeftModel
 from functools import lru_cache
 import os
 import time
+import threading
 import nltk
-import json
-import re
 
 try:
-    nltk.data.find('tokenizers/punkt')
+    nltk.data.find("tokenizers/punkt")
 except LookupError:
-    nltk.download('punkt')
+    nltk.download("punkt")
     try:
-        nltk.download('punkt_tab')
+        nltk.download("punkt_tab")
     except:
         pass
 
@@ -24,332 +23,164 @@ BASE_MODEL = os.getenv("BASE_MODEL", "facebook/nllb-200-distilled-600M")
 ADAPTER_EN_TO_NO = os.getenv("ADAPTER_EN_TO_NO", "entropy25/mt_en_no_oil")
 ADAPTER_NO_TO_EN = os.getenv("ADAPTER_NO_TO_EN", "entropy25/mt_no_en_oil")
 
-tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
+MAX_FILE_SIZE = 1024 * 1024
+MAX_TEXT_LENGTH = 10000
+BATCH_SIZE = 10
+NUM_BEAMS = 3
+MAX_LENGTH = 256
 
-print("Loading model with 8-bit quantization...")
-quantization_config = BitsAndBytesConfig(load_in_8bit=True)
+tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
 
 base_model = AutoModelForSeq2SeqLM.from_pretrained(
     BASE_MODEL,
-    quantization_config=quantization_config,
-    low_cpu_mem_usage=True
+    quantization_config=BitsAndBytesConfig(load_in_8bit=True),
+    low_cpu_mem_usage=True,
 )
 
-print("Loading adapters...")
 model = PeftModel.from_pretrained(base_model, ADAPTER_EN_TO_NO, adapter_name="en_to_no")
 model.load_adapter(ADAPTER_NO_TO_EN, adapter_name="no_to_en")
 model.eval()
 
-print("Loading terminology glossary...")
-try:
-    with open('glossary.json', 'r', encoding='utf-8') as f:
-        glossary_data = json.load(f)
-    
-    TERMINOLOGY_EN_TO_NO = {}
-    TERMINOLOGY_NO_TO_EN = {}
-    
-    for entry in glossary_data:
-        en_term = entry['en'].strip()
-        no_term = entry['no'].strip()
-        TERMINOLOGY_EN_TO_NO[en_term.lower()] = no_term
-        TERMINOLOGY_NO_TO_EN[no_term.lower()] = en_term
-    
-    print(f"Loaded {len(TERMINOLOGY_EN_TO_NO)} terminology entries")
-    
-except Exception as e:
-    print(f"Warning: Could not load glossary.json: {e}")
-    TERMINOLOGY_EN_TO_NO = {}
-    TERMINOLOGY_NO_TO_EN = {}
-
-COMPILED_PATTERNS_EN_TO_NO = {
-    term: re.compile(r'\b' + re.escape(term) + r'\b', re.IGNORECASE)
-    for term in TERMINOLOGY_EN_TO_NO.keys()
-}
+adapter_lock = threading.Lock()
 
-COMPILED_PATTERNS_NO_TO_EN = {
-    term: re.compile(r'\b' + re.escape(term) + r'\b', re.IGNORECASE)
-    for term in TERMINOLOGY_NO_TO_EN.keys()
+LANG_CONFIG = {
+    ("English", "Norwegian"): ("en_to_no", "nob_Latn"),
+    ("Norwegian", "English"): ("no_to_en", "eng_Latn"),
 }
 
-COMMON_ERRORS = {
-    "en_to_no": {
-        "mud weight": ["mudgevekten", "mudvekt", "slam vekt"],
-        "christmas tree": ["juletræet", "jule tre", "juletre"],
-        "permeability": ["permeabiliteten"],
-        "porosity": ["porøsiteten"],
-        "training": ["utdanning"],
-        "working pressure": ["arbeidstrykk"],
-    },
-    "no_to_en": {
-        "slamvekt": ["slam weight", "mudweight"],
-        "juletre": ["yule tree", "christmas-tree"],
-        "permeabilitet": ["permeabiliteten"],
-        "porøsitet": ["porøsiteten"],
-    }
+EXAMPLES_EN = {
+    "drilling_short": "Mud weight adjusted to 1.82 specific gravity at 3,247 meters depth.",
+    "drilling_long": "The drilling operation at well site A-15 encountered unexpected high-pressure zones at 3,247 meters depth, requiring immediate adjustment of mud weight from 1.65 to 1.82 specific gravity to maintain wellbore stability and prevent potential kicks.",
+    "reservoir_short": "Permeability is 250 millidarcy with 22 percent porosity.",
+    "reservoir_long": "The reservoir shows excellent permeability of 250 millidarcy and porosity of 22 percent based on core analysis, indicating significant hydrocarbon potential with estimated oil saturation of 65 percent.",
+    "subsea_short": "Christmas tree rated for 10,000 psi working pressure.",
+    "subsea_long": "The subsea production system consists of a vertical Christmas tree rated for 10,000 psi working pressure and 150 degrees Celsius temperature, equipped with redundant safety features including automatic shutdown valves and real-time pressure monitoring systems.",
+    "seismic_short": "Structural trap area estimated at 12 square kilometers.",
+    "seismic_long": "Seismic data confirms the presence of a structural trap with an estimated area of 12 square kilometers, and productivity tests show stabilized oil production of 3,400 barrels per day at optimization pressure of 2,100 psi.",
+    "safety_short": "H2S training required before site access.",
+    "safety_long": "Emergency response procedures require all personnel to complete H2S safety training before site access, with breathing apparatus and wind indicators positioned at designated muster points, and immediate evacuation protocols activated when gas detection exceeds 10 ppm concentration levels.",
 }
 
-COMPILED_ERRORS_EN_TO_NO = {}
-for source_term, error_variants in COMMON_ERRORS["en_to_no"].items():
-    COMPILED_ERRORS_EN_TO_NO[source_term] = [
-        re.compile(r'\b' + re.escape(variant) + r'\b', re.IGNORECASE)
-        for variant in error_variants
-    ]
-
-COMPILED_ERRORS_NO_TO_EN = {}
-for source_term, error_variants in COMMON_ERRORS["no_to_en"].items():
-    COMPILED_ERRORS_NO_TO_EN[source_term] = [
-        re.compile(r'\b' + re.escape(variant) + r'\b', re.IGNORECASE)
-        for variant in error_variants
-    ]
+EXAMPLES_NO = {
+    "drilling_short": "Slamvekt justert til 1,82 spesifikk tyngde ved 3 247 meters dybde.",
+    "drilling_long": "Boreoperasjonen ved brønnsted A-15 støtte på uventede høytrykksoner ved 3 247 meters dybde, noe som krevde umiddelbar justering av slamvekt fra 1,65 til 1,82 spesifikk tyngde for å opprettholde brønnborestabilitet og forhindre potensielle kicks.",
+    "reservoir_short": "Permeabilitet er 250 millidarcy med 22 prosent porøsitet.",
+    "reservoir_long": "Reservoaret viser utmerket permeabilitet på 250 millidarcy og porøsitet på 22 prosent basert på kjerneanalyse, noe som indikerer betydelig hydrokarbonpotensial med estimert oljemetning på 65 prosent.",
+    "subsea_short": "Juletre dimensjonert for 10 000 psi arbeidstrykk.",
+    "subsea_long": "Subsea produksjonssystemet består av et vertikalt juletre dimensjonert for 10 000 psi arbeidstrykk og 150 grader Celsius temperatur, utstyrt med redundante sikkerhetsfunksjoner inkludert automatiske nedstengningsventiler og sanntids trykkmonitorering.",
+    "seismic_short": "Strukturell felle estimert til 12 kvadratkilometer.",
+    "seismic_long": "Seismiske data bekrefter tilstedeværelsen av en strukturell felle med estimert areal på 12 kvadratkilometer, og produktivitetstester viser stabilisert oljeproduksjon på 3 400 fat per dag ved optimaliseringstrykk på 2 100 psi.",
+    "safety_short": "H2S-opplæring påkrevd før tilgang til området.",
+    "safety_long": "Nødprosedyrer krever at alt personell fullfører H2S-sikkerhetsopplæring før områdetilgang, med åndedrettsutstyr og vindindikatorer plassert ved utpekte samlingspunkter, og umiddelbare evakueringsprotokoller aktiveres når gassdeteksjon overskrider 10 ppm konsentrasjonsnivå.",
+}
 
-MAX_FILE_SIZE = 1024 * 1024
-MAX_TEXT_LENGTH = 10000
-BATCH_SIZE = 10
-NUM_BEAMS = 3
-MAX_LENGTH = 256
 
 @lru_cache(maxsize=512)
 def cached_sent_tokenize(text):
     return tuple(sent_tokenize(text))
 
-def fix_number_format(text, target_lang):
-    if target_lang == "Norwegian":
-        text = re.sub(r'(\d),(\d{3})', r'\1 \2', text)
-        text = re.sub(r'(\d)\.(\d{3})(?!\d)', r'\1 \2', text)
-        text = re.sub(r'(\d)\.(\d{1,2})(?=\s|$|[^\d])', r'\1,\2', text)
-    else:
-        text = re.sub(r'(\d)\s(\d{3})', r'\1,\2', text)
-        text = re.sub(r'(\d),(\d{1,2})(?=\s|$|[^\d])', r'\1.\2', text)
-    return text
-
-def find_source_terms_in_input(text, direction):
-    if direction == "en_to_no":
-        term_dict = TERMINOLOGY_EN_TO_NO
-    else:
-        term_dict = TERMINOLOGY_NO_TO_EN
-    
-    if not term_dict:
-        return []
-    
-    found_terms = []
-    text_lower = text.lower()
-    
-    for source_term in sorted(term_dict.keys(), key=len, reverse=True):
-        if source_term in text_lower:
-            target_term = term_dict[source_term]
-            found_terms.append((source_term, target_term))
-    
-    return found_terms
-
-def post_process_terminology(text, direction, found_terms, use_terminology):
-    if not use_terminology or not text:
-        return text
-    
-    if direction == "en_to_no":
-        compiled_patterns = COMPILED_PATTERNS_EN_TO_NO
-        compiled_errors = COMPILED_ERRORS_EN_TO_NO
-    else:
-        compiled_patterns = COMPILED_PATTERNS_NO_TO_EN
-        compiled_errors = COMPILED_ERRORS_NO_TO_EN
-    
-    result = text
-    
-    for source_term, target_term in found_terms:
-        def preserve_case(match):
-            original = match.group(0)
-            if original and original[0].isupper():
-                return target_term.capitalize()
-            return target_term.lower()
-        
-        if source_term in compiled_patterns:
-            result = compiled_patterns[source_term].sub(preserve_case, result)
-        
-        if source_term in compiled_errors:
-            for error_pattern in compiled_errors[source_term]:
-                result = error_pattern.sub(preserve_case, result)
-    
-    result = fix_number_format(result, "Norwegian" if direction == "en_to_no" else "English")
-    
-    return result
-
-def highlight_terminology(text, found_terms):
-    if not found_terms:
-        return text
-    
-    highlighted = text
-    for source_term, target_term in found_terms:
-        pattern = re.compile(r'\b(' + re.escape(target_term) + r')\b', re.IGNORECASE)
-        highlighted = pattern.sub(r'<mark style="background-color: #fff3cd; padding: 2px 4px; border-radius: 3px;">\1</mark>', highlighted)
-    
-    return highlighted
-
-def translate_core(text, source_lang, target_lang, use_terminology=True):
-    if not text.strip() or source_lang == target_lang:
-        return text, 0.0, []
-    
-    start_time = time.time()
-    
-    if source_lang == "English" and target_lang == "Norwegian":
-        model.set_adapter("en_to_no")
-        tgt_code = "nob_Latn"
-        direction = "en_to_no"
-    elif source_lang == "Norwegian" and target_lang == "English":
-        model.set_adapter("no_to_en")
-        tgt_code = "eng_Latn"
-        direction = "no_to_en"
-    else:
-        return "Unsupported language pair", 0.0, []
-    
-    found_terms = find_source_terms_in_input(text, direction)
-    
-    original_paragraphs = text.split('\n')
-    final_translated_paragraphs = []
-    
-    for paragraph in original_paragraphs:
+
+@lru_cache(maxsize=512)
+def translate_cached(text, source_lang, target_lang):
+    if source_lang == target_lang or not text.strip():
+        return text, 0.0
+
+    config = LANG_CONFIG.get((source_lang, target_lang))
+    if not config:
+        return "Unsupported language pair", 0.0
+
+    adapter_name, tgt_code = config
+    start = time.time()
+    device = next(model.parameters()).device
+    translated_paragraphs = []
+
+    for paragraph in text.split("\n"):
         if not paragraph.strip():
-            final_translated_paragraphs.append("")
+            translated_paragraphs.append("")
             continue
-        
+
         sentences = cached_sent_tokenize(paragraph)
-        paragraph_results = []
-        
+        results = []
+
         for i in range(0, len(sentences), BATCH_SIZE):
-            batch = sentences[i:i+BATCH_SIZE]
-            
+            batch = sentences[i : i + BATCH_SIZE]
             inputs = tokenizer(
                 batch,
                 return_tensors="pt",
                 padding=True,
                 truncation=True,
-                max_length=MAX_LENGTH
+                max_length=MAX_LENGTH,
             )
-            
-            if hasattr(model, 'device'):
-                inputs = {k: v.to(model.device) for k, v in inputs.items()}
-            
-            with torch.inference_mode():
-                outputs = model.generate(
-                    **inputs,
-                    forced_bos_token_id=tokenizer.convert_tokens_to_ids(tgt_code),
-                    max_length=MAX_LENGTH,
-                    num_beams=NUM_BEAMS,
-                    early_stopping=True
-                )
-            
-            batch_texts = tokenizer.batch_decode(outputs, skip_special_tokens=True)
-            paragraph_results.extend(batch_texts)
-        
-        final_translated_paragraphs.append(" ".join(paragraph_results))
-    
-    raw_translation = '\n'.join(final_translated_paragraphs)
-    corrected_translation = post_process_terminology(raw_translation, direction, found_terms, use_terminology)
-    elapsed_time = time.time() - start_time
-    
-    return corrected_translation, elapsed_time, found_terms
+            inputs = {k: v.to(device) for k, v in inputs.items()}
+
+            with adapter_lock:
+                model.set_adapter(adapter_name)
+                with torch.inference_mode():
+                    outputs = model.generate(
+                        **inputs,
+                        forced_bos_token_id=tokenizer.convert_tokens_to_ids(tgt_code),
+                        max_length=MAX_LENGTH,
+                        num_beams=NUM_BEAMS,
+                        early_stopping=True,
+                    )
+
+            results.extend(tokenizer.batch_decode(outputs, skip_special_tokens=True))
+
+        translated_paragraphs.append(" ".join(results))
+
+    return "\n".join(translated_paragraphs), time.time() - start
 
-@lru_cache(maxsize=512)
-def translate_cached(text, source_lang, target_lang, use_terminology):
-    result, elapsed, terms = translate_core(text, source_lang, target_lang, use_terminology)
-    return result, elapsed, len(terms)
 
-def translate(text, source_lang, target_lang, use_terminology):
+def translate(text, source_lang, target_lang):
+    if not text.strip():
+        return "", ""
+    if len(text) > MAX_TEXT_LENGTH:
+        return f"Error: Text too long (max {MAX_TEXT_LENGTH:,} characters)", ""
+
     try:
-        if len(text) > MAX_TEXT_LENGTH:
-            return f"Error: Text too long (max {MAX_TEXT_LENGTH:,} characters)", "", ""
-        
-        if not text.strip():
-            return "", "", ""
-        
-        result, elapsed, terms_count = translate_cached(text, source_lang, target_lang, use_terminology)
-        
-        terminology_status = f"{terms_count} terms enforced" if use_terminology and terms_count > 0 else "No terminology enforcement" if not use_terminology else "No terms found"
-        time_info = f"Completed in {elapsed:.2f}s | {terminology_status}"
-        
-        found_terms = find_source_terms_in_input(text, "en_to_no" if source_lang == "English" else "no_to_en")
-        highlighted_result = highlight_terminology(result, found_terms) if use_terminology else result
-        
-        return result, highlighted_result, time_info
-        
+        result, elapsed = translate_cached(text, source_lang, target_lang)
+        return result, f"Completed in {elapsed:.2f}s"
     except Exception as e:
-        return f"Translation error: {str(e)}", "", ""
+        return f"Translation error: {str(e)}", ""
+
 
 def swap_languages(src, tgt, input_txt, output_txt):
     return tgt, src, output_txt, input_txt
 
+
 def load_file(file):
     if file is None:
-        return "", "", ""
-    
+        return "", ""
     try:
-        if os.path.getsize(file.name) > MAX_FILE_SIZE:
-            return "Error: File too large (max 1MB)", "", ""
-        
-        with open(file.name, 'r', encoding='utf-8') as f:
-            content = f.read()
-            if len(content) > MAX_TEXT_LENGTH:
-                return f"Error: File content too long (max {MAX_TEXT_LENGTH:,} characters)", "", ""
-            return content, "", ""
-    except:
-        try:
-            with open(file.name, 'r', encoding='latin-1') as f:
-                content = f.read()
+        size = os.path.getsize(file.name)
+        if size > MAX_FILE_SIZE:
+            return "Error: File too large (max 1MB)", ""
+        for encoding in ("utf-8", "latin-1"):
+            try:
+                with open(file.name, "r", encoding=encoding) as f:
+                    content = f.read()
                 if len(content) > MAX_TEXT_LENGTH:
-                    return f"Error: File content too long (max {MAX_TEXT_LENGTH:,} characters)", "", ""
-                return content, "", ""
-        except Exception as e:
-            return f"Error reading file: {str(e)}", "", ""
-
-EXAMPLES_EN = {
-    "drilling_short": "Mud weight adjusted to 1.82 specific gravity at 3,247 meters depth.",
-    "drilling_long": "The drilling operation at well site A-15 encountered unexpected high-pressure zones at 3,247 meters depth, requiring immediate adjustment of mud weight from 1.65 to 1.82 specific gravity to maintain wellbore stability and prevent potential kicks.",
-    "reservoir_short": "Permeability is 250 millidarcy with 22 percent porosity.",
-    "reservoir_long": "The reservoir shows excellent permeability of 250 millidarcy and porosity of 22 percent based on core analysis, indicating significant hydrocarbon potential with estimated oil saturation of 65 percent.",
-    "subsea_short": "Christmas tree rated for 10,000 psi working pressure.",
-    "subsea_long": "The subsea production system consists of a vertical Christmas tree rated for 10,000 psi working pressure and 150 degrees Celsius temperature, equipped with redundant safety features including automatic shutdown valves and real-time pressure monitoring systems.",
-    "seismic_short": "Structural trap area estimated at 12 square kilometers.",
-    "seismic_long": "Seismic data confirms the presence of a structural trap with an estimated area of 12 square kilometers, and productivity tests show stabilized oil production of 3,400 barrels per day at optimization pressure of 2,100 psi.",
-    "safety_short": "H2S training required before site access.",
-    "safety_long": "Emergency response procedures require all personnel to complete H2S safety training before site access, with breathing apparatus and wind indicators positioned at designated muster points, and immediate evacuation protocols activated when gas detection exceeds 10 ppm concentration levels."
-}
-
-EXAMPLES_NO = {
-    "drilling_short": "Slamvekt justert til 1,82 spesifikk tyngde ved 3 247 meters dybde.",
-    "drilling_long": "Boreoperasjonen ved brønnsted A-15 støtte på uventede høytrykksoner ved 3 247 meters dybde, noe som krevde umiddelbar justering av slamvekt fra 1,65 til 1,82 spesifikk tyngde for å opprettholde brønnborestabilitet og forhindre potensielle kicks.",
-    "reservoir_short": "Permeabilitet er 250 millidarcy med 22 prosent porøsitet.",
-    "reservoir_long": "Reservoaret viser utmerket permeabilitet på 250 millidarcy og porøsitet på 22 prosent basert på kjerneanalyse, noe som indikerer betydelig hydrokarbonpotensial med estimert oljemetning på 65 prosent.",
-    "subsea_short": "Juletre dimensjonert for 10 000 psi arbeidstrykk.",
-    "subsea_long": "Subsea produksjonssystemet består av et vertikalt juletre dimensjonert for 10 000 psi arbeidstrykk og 150 grader Celsius temperatur, utstyrt med redundante sikkerhetsfunksjoner inkludert automatiske nedstengningsventiler og sanntids trykkmonitorering.",
-    "seismic_short": "Strukturell felle estimert til 12 kvadratkilometer.",
-    "seismic_long": "Seismiske data bekrefter tilstedeværelsen av en strukturell felle med estimert areal på 12 kvadratkilometer, og produktivitetstester viser stabilisert oljeproduksjon på 3 400 fat per dag ved optimaliseringstrykk på 2 100 psi.",
-    "safety_short": "H2S-opplæring påkrevd før tilgang til området.",
-    "safety_long": "Nødprosedyrer krever at alt personell fullfører H2S-sikkerhetsopplæring før områdetilgang, med åndedrettsutstyr og vindindikatorer plassert ved utpekte samlingspunkter, og umiddelbare evakueringsprotokoller aktiveres når gassdeteksjon overskrider 10 ppm konsentrasjonsnivå."
-}
-
-def get_examples(source_lang):
-    return EXAMPLES_EN if source_lang == "English" else EXAMPLES_NO
+                    return f"Error: File content too long (max {MAX_TEXT_LENGTH:,} characters)", ""
+                return content, ""
+            except UnicodeDecodeError:
+                continue
+    except (IOError, OSError) as e:
+        return f"Error reading file: {str(e)}", ""
 
-def update_example_buttons(source_lang):
-    examples = get_examples(source_lang)
-    return examples["drilling_short"]
 
 def get_example(key, source_lang):
-    examples = get_examples(source_lang)
+    examples = EXAMPLES_EN if source_lang == "English" else EXAMPLES_NO
     return examples[key]
 
+
 custom_css = """
 .gradio-container {
     max-width: 1100px !important;
     font-family: -apple-system, BlinkMacSystemFont, 'Segoe UI', sans-serif !important;
 }
-.main-container {
-    background: #f6f7f8 !important;
-    padding: 0 !important;
-    border-radius: 0 !important;
-}
 .translate-box {
     background: white !important;
     border-radius: 5px !important;
-    padding: 0 !important;
     box-shadow: 0 2px 4px rgba(0,0,0,0.08) !important;
     margin: 20px 0 !important;
 }
@@ -358,13 +189,6 @@ custom_css = """
     border-bottom: 1px solid #e8eaed !important;
     background: #fafafa !important;
 }
-.lang-selector {
-    border: none !important;
-    background: transparent !important;
-    font-size: 15px !important;
-    font-weight: 500 !important;
-    color: #333 !important;
-}
 .text-area textarea {
     border: none !important;
     font-size: 17px !important;
@@ -372,12 +196,6 @@ custom_css = """
     padding: 20px !important;
     min-height: 200px !important;
 }
-.swap-container {
-    display: flex !important;
-    align-items: center !important;
-    justify-content: center !important;
-    padding: 20px 0 !important;
-}
 .swap-btn {
     width: 44px !important;
     height: 44px !important;
@@ -388,11 +206,6 @@ custom_css = """
     box-shadow: 0 1px 3px rgba(0,0,0,0.1) !important;
     font-size: 18px !important;
     color: #0f6fff !important;
-    cursor: pointer !important;
-}
-.swap-btn:hover {
-    background: #f8f9fa !important;
-    border-color: #0f6fff !important;
 }
 .translate-btn {
     background: #ff8c00 !important;
@@ -402,10 +215,6 @@ custom_css = """
     font-size: 15px !important;
     font-weight: 500 !important;
     border-radius: 4px !important;
-    cursor: pointer !important;
-}
-.translate-btn:hover {
-    background: #e67e00 !important;
 }
 .time-info {
     text-align: center !important;
@@ -414,34 +223,12 @@ custom_css = """
     padding: 10px !important;
     font-style: italic !important;
 }
-.footer-info {
-    text-align: center !important;
-    color: #999 !important;
-    font-size: 13px !important;
-    padding: 20px !important;
-}
-.disclaimer {
-    background: #fff9e6 !important;
-    border-left: 4px solid #ff8c00 !important;
-    padding: 15px 20px !important;
-    margin: 20px 0 !important;
-    border-radius: 4px !important;
-    font-size: 13px !important;
-    color: #666 !important;
-}
 """
 
 with gr.Blocks(css=custom_css, theme=gr.themes.Default()) as demo:
-    
+
     gr.HTML("<div style='height: 20px'></div>")
-    
-    with gr.Row():
-        use_terminology = gr.Checkbox(
-            label="Enable Terminology Enforcement",
-            value=True,
-            info=f"Post-processing with {len(TERMINOLOGY_EN_TO_NO)} oil & gas terms"
-        )
-    
+
     with gr.Row():
         with gr.Column(scale=1):
             with gr.Group(elem_classes="translate-box"):
@@ -451,8 +238,7 @@ with gr.Blocks(css=custom_css, theme=gr.themes.Default()) as demo:
                         value="English",
                         show_label=False,
                         container=False,
-                        elem_classes="lang-selector",
-                        scale=1
+                        scale=1,
                     )
                 input_text = gr.Textbox(
                     placeholder="Type to translate",
@@ -460,13 +246,12 @@ with gr.Blocks(css=custom_css, theme=gr.themes.Default()) as demo:
                     lines=8,
                     max_lines=20,
                     container=False,
-                    elem_classes="text-area"
+                    elem_classes="text-area",
                 )
-        
+
         with gr.Column(scale=0, min_width=100):
-            with gr.Row(elem_classes="swap-container"):
-                swap_btn = gr.Button("⇄", elem_classes="swap-btn")
-        
+            swap_btn = gr.Button("⇄", elem_classes="swap-btn")
+
         with gr.Column(scale=1):
             with gr.Group(elem_classes="translate-box"):
                 with gr.Row(elem_classes="lang-header"):
@@ -475,10 +260,9 @@ with gr.Blocks(css=custom_css, theme=gr.themes.Default()) as demo:
                         value="Norwegian",
                         show_label=False,
                         container=False,
-                        elem_classes="lang-selector",
-                        scale=1
+                        scale=1,
                     )
-                output_text_plain = gr.Textbox(
+                output_text = gr.Textbox(
                     placeholder="Translation",
                     show_label=False,
                     lines=8,
@@ -486,94 +270,64 @@ with gr.Blocks(css=custom_css, theme=gr.themes.Default()) as demo:
                     container=False,
                     elem_classes="text-area",
                     interactive=False,
-                    visible=False
-                )
-                output_text_html = gr.HTML(
-                    value="<div style='padding: 20px; min-height: 200px; font-size: 17px; line-height: 1.7;'>Translation</div>"
                 )
-    
+
     with gr.Row():
         translate_btn = gr.Button("Translate", variant="primary", elem_classes="translate-btn", size="lg")
-    
+
     with gr.Row():
-        time_display = gr.Textbox(
-            show_label=False,
-            container=False,
-            interactive=False,
-            elem_classes="time-info"
-        )
-    
-    gr.HTML("<div class='footer-info'>Oil & Gas Translation • English ↔ Norwegian • Terminology Highlighting</div>")
-    
+        time_display = gr.Textbox(show_label=False, container=False, interactive=False, elem_classes="time-info")
+
+    gr.HTML("<div style='text-align:center;color:#999;font-size:13px;padding:20px'>Oil & Gas Translation — English ↔ Norwegian</div>")
+
     with gr.Accordion("Example Sentences", open=True):
         with gr.Row():
-            example_text = gr.Textbox(
-                value=EXAMPLES_EN["drilling_short"],
-                label="",
-                lines=3,
-                max_lines=5,
-                show_copy_button=True
-            )
+            example_text = gr.Textbox(value=EXAMPLES_EN["drilling_short"], label="", lines=3, show_copy_button=True)
             use_example_btn = gr.Button("Use This Example", variant="primary", size="sm")
-        
+
         with gr.Row():
-            btn1 = gr.Button("Drilling Short", size="sm")
-            btn2 = gr.Button("Drilling Long", size="sm")
-            btn3 = gr.Button("Reservoir Short", size="sm")
-            btn4 = gr.Button("Reservoir Long", size="sm")
-            btn5 = gr.Button("Subsea Short", size="sm")
-        
+            for key, label in [
+                ("drilling_short", "Drilling Short"),
+                ("drilling_long", "Drilling Long"),
+                ("reservoir_short", "Reservoir Short"),
+                ("reservoir_long", "Reservoir Long"),
+                ("subsea_short", "Subsea Short"),
+            ]:
+                gr.Button(label, size="sm").click(
+                    lambda sl, k=key: get_example(k, sl), inputs=[source_lang], outputs=example_text
+                )
+
         with gr.Row():
-            btn6 = gr.Button("Subsea Long", size="sm")
-            btn7 = gr.Button("Seismic Short", size="sm")
-            btn8 = gr.Button("Seismic Long", size="sm")
-            btn9 = gr.Button("Safety Short", size="sm")
-            btn10 = gr.Button("Safety Long", size="sm")
-        
-        btn1.click(lambda sl: get_example("drilling_short", sl), inputs=[source_lang], outputs=example_text)
-        btn2.click(lambda sl: get_example("drilling_long", sl), inputs=[source_lang], outputs=example_text)
-        btn3.click(lambda sl: get_example("reservoir_short", sl), inputs=[source_lang], outputs=example_text)
-        btn4.click(lambda sl: get_example("reservoir_long", sl), inputs=[source_lang], outputs=example_text)
-        btn5.click(lambda sl: get_example("subsea_short", sl), inputs=[source_lang], outputs=example_text)
-        btn6.click(lambda sl: get_example("subsea_long", sl), inputs=[source_lang], outputs=example_text)
-        btn7.click(lambda sl: get_example("seismic_short", sl), inputs=[source_lang], outputs=example_text)
-        btn8.click(lambda sl: get_example("seismic_long", sl), inputs=[source_lang], outputs=example_text)
-        btn9.click(lambda sl: get_example("safety_short", sl), inputs=[source_lang], outputs=example_text)
-        btn10.click(lambda sl: get_example("safety_long", sl), inputs=[source_lang], outputs=example_text)
-        
+            for key, label in [
+                ("subsea_long", "Subsea Long"),
+                ("seismic_short", "Seismic Short"),
+                ("seismic_long", "Seismic Long"),
+                ("safety_short", "Safety Short"),
+                ("safety_long", "Safety Long"),
+            ]:
+                gr.Button(label, size="sm").click(
+                    lambda sl, k=key: get_example(k, sl), inputs=[source_lang], outputs=example_text
+                )
+
         use_example_btn.click(fn=lambda x: x, inputs=example_text, outputs=input_text)
-    
+
     with gr.Accordion("Upload Text File", open=False):
-        file_input = gr.File(
-            label="Upload a .txt file to translate (max 1MB)",
-            file_types=[".txt"],
-            type="filepath"
-        )
-
-    gr.HTML(f"""
-    <div class='disclaimer'>
-        <strong>Terminology Enforcement:</strong> {len(TERMINOLOGY_EN_TO_NO)} oil & gas terms with automatic highlighting
-        <br>
-        <strong>Privacy & Compliance:</strong> Local inference ensures GDPR compliance
-        <br>
-        <strong>Technical Features:</strong> Optimized batch processing with pre-compiled regex patterns
-    </div>
-    """)
-    
-    source_lang.change(fn=update_example_buttons, inputs=[source_lang], outputs=[example_text])
-    
+        file_input = gr.File(label="Upload a .txt file (max 1MB)", file_types=[".txt"], type="filepath")
+
     translate_btn.click(
         fn=translate,
-        inputs=[input_text, source_lang, target_lang, use_terminology],
-        outputs=[output_text_plain, output_text_html, time_display]
+        inputs=[input_text, source_lang, target_lang],
+        outputs=[output_text, time_display],
     )
-    
+
     swap_btn.click(
         fn=swap_languages,
-        inputs=[source_lang, target_lang, input_text, output_text_plain],
-        outputs=[source_lang, target_lang, input_text, output_text_plain]
+        inputs=[source_lang, target_lang, input_text, output_text],
+        outputs=[source_lang, target_lang, input_text, output_text],
     )
-    
-    file_input.change(fn=load_file, inputs=file_input, outputs=[input_text, output_text_html, time_display])
+
+    file_input.change(fn=load_file, inputs=file_input, outputs=[input_text, time_display])
+
+    source_lang.change(fn=lambda sl: get_example("drilling_short", sl), inputs=[source_lang], outputs=[example_text])
 
 demo.queue().launch()