removed some debug functions and preload. also made auto political text generation to be capped in 1 sentence and lessen max_new_token to 40.

Dockerfile

@@ -15,4 +15,4 @@ RUN python -c "import nltk; nltk.download('punkt'); nltk.download('punkt_tab')"
 RUN python -c "from transformers import pipeline; pipeline('text-generation', model='gpt2')"
 
 COPY --chown=user . /app
-CMD ["gunicorn", "-w", "2", "--preload", "--capture-output", "--log-level", "debug", "-b", "0.0.0.0:7860", "app:app"]
+CMD ["gunicorn", "-w", "1", "-b", "0.0.0.0:7860", "app:app"]

app.py

@@ -13,25 +13,16 @@ app = Flask(__name__)
 tokenizer = None
 model = None
 generator = None
-startup_errors = []
-
-try:
-    print("Loading sentiment model...")
-    tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=TOKEN)
-    model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME, token=TOKEN)
-    model.eval()
-    print("Sentiment model loaded!")
-except Exception as e:
-    startup_errors.append({"stage": "sentiment_model", "error": traceback.format_exc()})
-    print(f"Sentiment model failed: {e}")
-
-try:
-    print("Loading generation model...")
-    generator = pipeline('text-generation', model='gpt2', device=-1)
-    print("Generation model loaded!")
-except Exception as e:
-    startup_errors.append({"stage": "generation_model", "error": traceback.format_exc()})
-    print(f"Generation model failed: {e}")
+
+print("Loading sentiment model...")
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, token=TOKEN)
+model = AutoModelForSequenceClassification.from_pretrained(MODEL_NAME, token=TOKEN)
+model.eval()
+print("Sentiment model loaded!")
+
+print("Loading generation model...")
+generator = pipeline('text-generation', model='gpt2', device=-1)
+print("Generation model loaded!")
 
 POLITICAL_SEEDS = [
     "The Senate voted",
@@ -78,9 +69,13 @@ def is_political(text, seed):
 def trim_to_sentences(text, seed):
     body = text[len(seed):].strip() if text.startswith(seed) else text.strip()
     full = (seed + ' ' + body).strip()
-    last_punct = max(full.rfind('.'), full.rfind('!'), full.rfind('?'))
-    if last_punct > len(seed):
-        full = full[:last_punct + 1]
+    first_punct = -1
+    for i, ch in enumerate(full):
+        if i > len(seed) and ch in '.!?':
+            first_punct = i
+            break
+    if first_punct != -1:
+        full = full[:first_punct + 1]
     return full.strip()
 
 def validate_and_clean_text(text):
@@ -118,27 +113,6 @@ def get_sentiment_score(text):
 def get_sentiment_label(score):
     return {-1: "Negative", 0: "Neutral", 1: "Positive"}.get(score, "Unknown")
 
-@app.route('/debug')
-def debug():
-    info = {
-        'startup_errors': startup_errors,
-        'model_loaded': model is not None,
-        'tokenizer_loaded': tokenizer is not None,
-        'generator_loaded': generator is not None,
-        'MODEL_NAME': os.environ.get('MODEL_NAME', 'NOT SET'),
-        'HF_TOKEN_set': bool(os.environ.get('HF_TOKEN')),
-    }
-    try:
-        info['nltk_test'] = sent_tokenize("Hello world. This is a test.")
-    except Exception:
-        info['nltk_error'] = traceback.format_exc()
-    if model is not None:
-        try:
-            info['sentiment_test'] = get_sentiment_score("This is a test.")
-        except Exception:
-            info['sentiment_error'] = traceback.format_exc()
-    return jsonify(info)
-
 @app.route('/generate', methods=['POST'])
 def generate():
     if generator is None:
@@ -150,7 +124,7 @@ def generate():
             seed = random.choice(POLITICAL_SEEDS)
             output = generator(
                 seed,
-                max_new_tokens=80,
+                max_new_tokens=40,
                 temperature=0.95,
                 top_p=0.92,
                 repetition_penalty=1.35,
@@ -162,7 +136,7 @@ def generate():
                 result_text = trimmed
                 break
         return jsonify({'text': result_text or trimmed})
-    except Exception as e:
+    except:
         return jsonify({'error': f'Generation failed: {traceback.format_exc()}'}), 500
 
 @app.route('/predict', methods=['POST'])
@@ -185,7 +159,7 @@ def predict():
             'original_text': text,
             'cleaned_text': cleaned_text
         })
-    except Exception as e:
+    except:
         return jsonify({'error': f'An error occurred: {traceback.format_exc()}'}), 500
 
 def load_base64_from_file(filename):

requirements.txt

@@ -2,7 +2,5 @@ gunicorn==23.0.0
 Flask==3.0.3
 transformers==4.46.3
 torch==2.4.1
-accelerate==0.34.2
-numpy==1.24.4
 python-dotenv==1.0.1
 nltk==3.9.1

templates/index.html

@@ -308,7 +308,6 @@
             </div>
             <div class="error hide" id="error"></div>
         </div>
-        <p class="startup-note">⏳ Auto generating political text may take up to 30 seconds on free hosting.</p>
     </div>
 
     <script>