Update app.py

app.py

@@ -1,5 +1,7 @@
+import os
 import re
 from typing import List, Tuple
+
 import gradio as gr
 from transformers import pipeline
 
@@ -8,6 +10,9 @@ from transformers import pipeline
 # -----------------------------
 
 MODEL_ID = "fakespot-ai/roberta-base-ai-text-detection-v1"
+
+# If you’re on CPU-only Space and want to be explicit, uncomment device=-1
+# clf = pipeline("text-classification", model=MODEL_ID, device=-1)
 clf = pipeline("text-classification", model=MODEL_ID)
 
 def clean_text(s: str) -> str:
@@ -31,20 +36,22 @@ def chunk_text(text: str, max_words: int = 300) -> List[str]:
 def detect_ai(text: str) -> Tuple[str, float, str]:
     """
     Returns (label, score_float, explanation)
+    - label: "AI" or "Human"
+    - score_float: mean AI likelihood in [0,1]
+    - explanation: short narrative with a few heuristic cues
     """
     if not text or not text.strip():
         return "—", 0.0, "Please paste some text to analyze."
 
     chunks = [clean_text(c) for c in chunk_text(text, max_words=300)]
-    # Run model in batch for speed
+
+    # Batch for speed and lower overhead
     preds = clf(chunks)
 
-    # Aggregate: average confidence across chunks for the detected class
-    # The model returns a label per chunk; we map AI=1, Human=0 and average
+    # Aggregate AI likelihood: if a chunk label is 'AI', use score; if 'Human', use (1-score)
     ai_probs = []
     for p in preds:
-        # Some models return 'AI'/'Human' labels; normalize
-        label = p.get("label", "").upper()
+        label = str(p.get("label", "")).upper()
         score = float(p.get("score", 0.0))
         ai_prob = score if label.startswith("AI") else (1.0 - score)
         ai_probs.append(ai_prob)
@@ -52,9 +59,7 @@ def detect_ai(text: str) -> Tuple[str, float, str]:
     mean_ai = sum(ai_probs) / len(ai_probs)
     label = "AI" if mean_ai >= 0.5 else "Human"
 
-    # Lightweight heuristic explanation (no extra LLM needed)
     explanation = build_explanation(text, mean_ai, len(chunks))
-
     return label, float(mean_ai), explanation
 
 def build_explanation(text: str, ai_prob: float, n_chunks: int) -> str:
@@ -63,7 +68,10 @@ def build_explanation(text: str, ai_prob: float, n_chunks: int) -> str:
     words = [w for w in words if w.strip()]
     sentences = [s for s in sentences if s.strip()]
 
-    avg_len = (sum(len(s.split()) for s in sentences) / max(1, len(sentences))) if sentences else 0
+    avg_len = (
+        sum(len(s.split()) for s in sentences) / max(1, len(sentences))
+        if sentences else 0
+    )
     vocab = set(w.lower() for w in words)
     ttr = len(vocab) / max(1, len(words))  # type-token ratio
 
@@ -92,7 +100,7 @@ def build_explanation(text: str, ai_prob: float, n_chunks: int) -> str:
     return (
         f"Overall this text is estimated to be {ai_prob:.2%} likely AI-generated. "
         f"Notable cues: " + "; ".join(cues) + ". "
-        "Remember: detectors can be wrong—use results as a hint, not proof."
+        "Reminder: detectors can be wrong—use results as a hint, not proof."
     )
 
 # -----------------------------
@@ -100,21 +108,31 @@ def build_explanation(text: str, ai_prob: float, n_chunks: int) -> str:
 # -----------------------------
 
 with gr.Blocks(title="AI Text Detector") as demo:
-    gr.Markdown("## 🕵️ AI Text Detector (Simple)\nPaste text and get an approximate AI-likeness score.\n"
-                "> Model: `fakespot-ai/roberta-base-ai-text-detection-v1`")
+    gr.Markdown(
+        "## 🕵️ AI Text Detector (Simple)\n"
+        "Paste text and get an approximate AI-likeness score.\n\n"
+        "> Model: `fakespot-ai/roberta-base-ai-text-detection-v1`"
+    )
 
     with gr.Row():
         inp = gr.Textbox(label="Input Text", lines=14, placeholder="Paste your text here...")
+
     with gr.Row():
         label_out = gr.Label(label="Predicted Class")
         score_out = gr.Slider(label="AI Likelihood", minimum=0.0, maximum=1.0, step=0.001, interactive=False)
+
     explain = gr.Textbox(label="Explanation", lines=6)
 
-    def _run(t):
+    def _run(t: str):
         label, score, expl = detect_ai(t)
+        # gr.Label expects a dict of {class_name: confidence} for pretty display
         return {label_out: {label: 1.0}, score_out: score, explain: expl}
 
     gr.Button("Analyze").click(_run, inputs=inp, outputs=[label_out, score_out, explain])
 
 if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=int(os.getenv("PORT", 7860)))
+    # For Spaces, PORT is provided by the environment
+    demo.queue(concurrency_count=1).launch(
+        server_name="0.0.0.0",
+        server_port=int(os.getenv("PORT", 7860))
+    )