fix translation

app.py

@@ -4,7 +4,6 @@ import libsql_experimental as libsql
 from qdrant_client import QdrantClient, models
 from sentence_transformers import SentenceTransformer
 import time
-import gradio as gr
 from transformers import AutoTokenizer, AutoModelForSeq2SeqLM, pipeline
 from langdetect import detect, DetectorFactory
 
@@ -12,45 +11,54 @@ DetectorFactory.seed = 0
 
 # --- SETUP MODELLO UNIVERSALE (NLLB) ---
 print("Caricamento modello NLLB (Any -> Any)...")
-# NLLB-200 Distilled: Ottimo compromesso qualità/peso (600M params)
 model_name = "facebook/nllb-200-distilled-600M"
 
-# Carichiamo tokenizzatore e modello
 tokenizer = AutoTokenizer.from_pretrained(model_name)
 model_trans = AutoModelForSeq2SeqLM.from_pretrained(model_name)
-
-# Creiamo una pipeline di traduzione generica
-# Nota: NLLB usa codici lingua specifici (es. 'eng_Latn', 'ita_Latn')
 translator_pipe = pipeline("translation", model=model_trans, tokenizer=tokenizer, device=-1)
 
-# Mappa semplice per convertire i codici lingua
-LANG_CODES = {
+# Mappa per convertire i codici semplici (langdetect) nei codici NLLB
+# Se il libro è in una lingua non in lista, useremo 'eng_Latn' come fallback
+CODE_TO_NLLB = {
     "en": "eng_Latn",
-    "it": "ita_Latn"
+    "it": "ita_Latn",
+    "fr": "fra_Latn",
+    "es": "spa_Latn",
+    "de": "deu_Latn",
+    "pt": "por_Latn",
+    "nl": "nld_Latn",
+    "ru": "rus_Cyrl",
+    "zh": "zho_Hans",
+    "ja": "jpn_Jpan"
 }
 
-def translate_wrapper(text, target_lang_code):
+def translate_wrapper(text, source_lang_simple, target_lang_simple):
     """
     Funzione helper per tradurre con NLLB.
     """
     try:
-        # Recupera il codice NLLB
-        nllb_code = LANG_CODES.get(target_lang_code, "eng_Latn")
-        
-        target_lang_id = tokenizer.convert_tokens_to_ids(nllb_code)
+        # Convertiamo i codici (es. 'en' -> 'eng_Latn')
+        src_nllb = CODE_TO_NLLB.get(source_lang_simple, "eng_Latn") # Default Inglese se sconosciuto
+        tgt_nllb = CODE_TO_NLLB.get(target_lang_simple, "eng_Latn")
+
+        # Diciamo al tokenizer qual è la lingua di partenza
+        tokenizer.src_lang = src_nllb
+
+        # Otteniamo l'ID della lingua di arrivo
+        target_lang_id = tokenizer.convert_tokens_to_ids(tgt_nllb)
         
+        # Eseguiamo la traduzione
         output = translator_pipe(
             text, 
-            forced_bos_token_id=target_lang_id
+            forced_bos_token_id=target_lang_id, # Forza lingua output
         )
         return output[0]['translation_text']
         
     except Exception as e:
         print(f"Errore NLLB: {e}")
-        # In caso di errore restituisce il testo originale con un avviso visibile in console
         return text
 
-# --- SETUP ---
+# --- SETUP DATABASES & EMBEDDER ---
 model = SentenceTransformer("sentence-transformers/paraphrase-multilingual-MiniLM-L12-v2", device="cpu")
 
 QDRANT_URL = os.environ.get("QDRANT_URL")
@@ -59,10 +67,7 @@ TURSO_URL = os.environ.get("TURSO_URL")
 TURSO_TOKEN = os.environ.get("TURSO_TOKEN")
 
 try:
-    client = QdrantClient(
-        url=QDRANT_URL,
-        api_key=QDRANT_API_KEY
-    )
+    client = QdrantClient(url=QDRANT_URL, api_key=QDRANT_API_KEY)
 except Exception as e:
     print(f"Errore Qdrant: {e}")
 
@@ -72,61 +77,26 @@ def get_turso_conn():
 COLLECTION_NAME = "books_collection"
 VECTOR_SIZE = 256
 
-def render_results_from_ids(ids, scores, target_lang="en"):
-    """
-    target_lang: 'en' (Inglese) o 'it' (Italiano)
-    """
-    if not ids: return "Nessun risultato trovato."
-
-# --- JAVASCRIPT PER FORZARE DARK MODE ---
+# --- JAVASCRIPT & CSS ---
 JS_FORCE_DARK = """
 function() {
     document.body.classList.add('dark');
 }
 """
 
-# --- CSS GLOBALE ---
 GLOBAL_CSS = """
-/* Sfondo pagina e testo */
-body, .gradio-container {
-    background-color: #0b0f19 !important;
-    color: #e5e7eb !important;
-}
-
-/* Animazione Spinner */
-.loader {
-  border: 6px solid #374151;
-  border-radius: 50%;
-  border-top: 6px solid #3b82f6;
-  border-bottom: 6px solid #ef4444;
-  width: 40px;
-  height: 40px;
-  -webkit-animation: spin 1s linear infinite;
-  animation: spin 1s linear infinite;
-  margin: 0 auto;
-}
+body, .gradio-container { background-color: #0b0f19 !important; color: #e5e7eb !important; }
+.loader { border: 6px solid #374151; border-radius: 50%; border-top: 6px solid #3b82f6; border-bottom: 6px solid #ef4444; width: 40px; height: 40px; animation: spin 1s linear infinite; margin: 0 auto; }
 @keyframes spin { 0% { transform: rotate(0deg); } 100% { transform: rotate(360deg); } }
-
-/* Bottoni Dataset */
-#book_cards button {
-    background-color: #1f2937 !important;
-    color: #e5e7eb !important;
-    border: 1px solid #374151 !important;
-}
-#book_cards button:hover {
-    background-color: #374151 !important;
-    border-color: #60a5fa !important;
-}
+#book_cards button { background-color: #1f2937 !important; color: #e5e7eb !important; border: 1px solid #374151 !important; }
+#book_cards button:hover { background-color: #374151 !important; border-color: #60a5fa !important; }
 #book_cards .text-sm { color: #9ca3af !important; }
-
-/* Stile Card Risultati */
 .card-force-dark { color: #e5e7eb !important; }
 .card-force-dark h3 { color: #60a5fa !important; margin-top: 0 !important; }
 .card-force-dark p, .card-force-dark b, .card-force-dark span { color: #d1d5db !important; }
 .card-force-dark summary { color: #fbbf24 !important; }
 """
 
-# HTML dello spinner
 LOADING_HTML = """
 <div style="display: flex; justify-content: center; align-items: center; height: 100px; width: 100%; flex-direction: column;">
     <div class="loader"></div>
@@ -149,10 +119,8 @@ def render_results_from_ids(ids, scores, target_lang="en"):
         cursor = conn.execute(sql_query, tuple(ids))
         rows = cursor.fetchall()
         books_map = {row[0]: row for row in rows}
-
         for uid in ids:
-            if uid in books_map:
-                ordered_books.append(books_map[uid])
+            if uid in books_map: ordered_books.append(books_map[uid])
     except Exception as e:
         return f"Errore Database Turso: {str(e)}"
     finally:
@@ -161,37 +129,37 @@ def render_results_from_ids(ids, scores, target_lang="en"):
     html_output = "<div style='font-family: sans-serif; gap: 10px; display: flex; flex-direction: column;'>"
     for row in ordered_books:
         score = scores.get(row[0], 0.0)
-
         autore_clean = str(row[2]).replace('"', '').replace("[","").replace("]", "")
-
         original_summary = row[5]
+        
         trama_display = "No summary available."
         lang_label = ""
 
-        # --- LOGICA DI TRADUZIONE INTELLIGENTE ---
+        # --- LOGICA TRADUZIONE ---
         if original_summary and len(original_summary) > 10:
             try:
-                # Rileva lingua originale
+                # Rileva lingua ORIGINALE (Source)
                 try:
-                    detected_lang = detect(original_summary)
+                    detected_src = detect(original_summary)
                 except:
-                    detected_lang = "unknown"
+                    detected_src = "en" # Fallback se fallisce il rilevamento
 
-                # Se la lingua è già quella target, non fare nulla
-                if detected_lang == target_lang:
+                # Se la lingua è già quella target, ok
+                if detected_src == target_lang:
                     trama_display = original_summary
                     lang_label = ""
                 
-                # Altrimenti traduci
+                # Altrimenti traduci passando Source -> Target
                 else:
-                    trama_display = translate_wrapper(original_summary, target_lang)
-                    lang_label = f"(Translated to {target_lang.upper()})"
+                    trama_display = translate_wrapper(original_summary, detected_src, target_lang)
+                    lang_label = f"(Translated from {detected_src.upper()})"
             
             except Exception as e:
+                print(f"Err render: {e}")
                 trama_display = original_summary
         else:
             trama_display = original_summary
-        # -----------------------------------------
+        # -------------------------
 
         label_trama = "Leggi Trama" if target_lang == 'it' else "Read Summary"
 
@@ -216,11 +184,9 @@ def render_results_from_ids(ids, scores, target_lang="en"):
 def search_free_text_animated(query_text, max_results, target_lang):
     yield gr.update(visible=True), gr.update(visible=False)
     time.sleep(0.2)
-
     if not query_text:
         yield gr.update(visible=False), "Inserisci una richiesta!"
         return
-
     try:
         vec = model.encode(f"{query_text}")[:VECTOR_SIZE]
         hits_response = client.query_points(
@@ -232,60 +198,43 @@ def search_free_text_animated(query_text, max_results, target_lang):
         hits = hits_response.points
         ids = [hit.id for hit in hits]
         scores = {hit.id: hit.score for hit in hits}
-
         final_html = render_results_from_ids(ids, scores, target_lang)
-
         yield gr.update(visible=False), gr.update(value=final_html, visible=True)
-
     except Exception as e:
         yield gr.update(visible=False), f"Errore: {e}"
 
-
 def find_book_cards_animated(partial_title):
-    # FASE 1: Spinner ON, Dataset OFF
     yield gr.update(visible=True), gr.update(visible=False), [], gr.update(value="", visible=False)
     time.sleep(0.3)
-
     if not partial_title or len(partial_title) < 2:
         yield gr.update(visible=False), gr.update(samples=[], visible=False), [], gr.update(visible=False)
         return
-
     conn = None
     try:
         conn = get_turso_conn()
         query = f"SELECT id, title, author, year FROM books WHERE title LIKE '%{partial_title}%' LIMIT 10"
         rows = conn.execute(query).fetchall()
-
         card_data = [[str(row[1]), str(row[2]).replace('"', '').replace("'", "").replace("[","").replace("]",""), str(row[3]).split('.')[0]] for row in rows]
         full_data_state = [{"id": row[0], "title": row[1]} for row in rows]
-
-        # FASE 2: Spinner OFF, Dataset ON
         yield gr.update(visible=False), gr.update(samples=card_data, visible=True), full_data_state, gr.update(visible=False)
-
     except Exception as e:
-        print(f"Error: {e}")
         yield gr.update(visible=False), gr.update(visible=False), [], gr.update(visible=False)
     finally:
         if conn: conn.close()
 
-
 def on_card_click_animated(selected_index, books_state_list, max_results, target_lang):
     yield gr.update(visible=True), gr.update(visible=False)
     time.sleep(0.2)
-
     if selected_index >= len(books_state_list):
         yield gr.update(visible=False), "Errore selezione."
         return
-
     book_obj = books_state_list[selected_index]
     source_id = book_obj["id"]
-
     try:
         points = client.retrieve(collection_name=COLLECTION_NAME, ids=[source_id], with_vectors=True)
         if not points:
             yield gr.update(visible=False), "ID non trovato."
             return
-
         existing_vector = points[0].vector
         hits_response = client.query_points(
             collection_name=COLLECTION_NAME,
@@ -297,28 +246,16 @@ def on_card_click_animated(selected_index, books_state_list, max_results, target
         ids = [hit.id for hit in hits_response.points]
         scores = {hit.id: hit.score for hit in hits_response.points}
         final_html = render_results_from_ids(ids, scores, target_lang)
-
         yield gr.update(visible=False), gr.update(value=final_html, visible=True)
-
     except Exception as e:
         yield gr.update(visible=False), f"Errore Backend: {e}"
 
-def clear_search_tab():
-    """Pulisce il Tab Ricerca per Trama quando si passa all'altro."""
-    return gr.update(value=""), gr.update(value="", visible=False)
-
-def clear_find_tab():
-    """Pulisce il Tab Mi è piaciuto quando si passa all'altro."""
-    return (
-        gr.update(value=""),           # txt_title
-        gr.update(visible=False),      # cards_view (Dataset)
-        gr.update(value="", visible=False) # out_results
-    )
+def clear_search_tab(): return gr.update(value=""), gr.update(value="", visible=False)
+def clear_find_tab(): return gr.update(value=""), gr.update(visible=False), gr.update(value="", visible=False)
 
 # --- INTERFACCIA ---
 with gr.Blocks(theme=gr.themes.Ocean(), css=GLOBAL_CSS, js=JS_FORCE_DARK) as demo:
     gr.Markdown("# 📚 AI Book Finder")
-
     books_state = gr.State([])
 
     with gr.Row():
@@ -326,61 +263,31 @@ with gr.Blocks(theme=gr.themes.Ocean(), css=GLOBAL_CSS, js=JS_FORCE_DARK) as dem
         lang_choice = gr.Dropdown(choices=["en", "it"], value="en", label="Lingua Trama / Summary Language")
 
     with gr.Tabs() as tabs: 
-        
-        # TAB 1: Ricerca Libera
-        with gr.Tab("🔎 Ricerca per Trama") as tab_trama: # Assegna variabile
+        with gr.Tab("🔎 Ricerca per Trama") as tab_trama:
             with gr.Row():
-                txt_input = gr.Textbox(placeholder="Descrivi la trama, l'atmosfera o le emozioni che cerchi...", show_label=False, scale=4)
+                txt_input = gr.Textbox(placeholder="Descrivi la trama...", show_label=False, scale=4)
                 btn_search = gr.Button("Cerca", variant="primary", scale=1)
 
-        # TAB 2: Ricerca per Libro
-        with gr.Tab("📖 Mi è piaciuto...") as tab_libro: # Assegna variabile
+        with gr.Tab("📖 Mi è piaciuto...") as tab_libro:
             with gr.Row():
-                txt_title = gr.Textbox(placeholder="Scrivi il titolo, anche parziale", show_label=False, scale=4)
+                txt_title = gr.Textbox(placeholder="Scrivi il titolo...", show_label=False, scale=4)
                 btn_find = gr.Button("Trova", scale=1)
-
             loader_cards = gr.HTML(value=LOADING_HTML, visible=False)
-
-            # DATASET
-            cards_view = gr.Dataset(
-                elem_id="book_cards",
-                label="Seleziona il libro corretto:",
-                components=[gr.Textbox(visible=False), gr.Textbox(visible=False), gr.Textbox(visible=False)],
-                headers=["Titolo", "Autore", "Anno"],
-                samples=[],
-                visible=False,
-                type="index"
-            )
+            cards_view = gr.Dataset(elem_id="book_cards", label="Seleziona il libro:", components=[gr.Textbox(visible=False), gr.Textbox(visible=False), gr.Textbox(visible=False)], headers=["Titolo", "Autore", "Anno"], samples=[], visible=False, type="index")
 
     loader_results = gr.HTML(value=LOADING_HTML, visible=False)
     out_results = gr.HTML(label="Consigli", visible=True)
 
-    # --- EVENTI DI RICERCA ---
     btn_search.click(fn=search_free_text_animated, inputs=[txt_input, num_results, lang_choice], outputs=[loader_results, out_results])
     txt_input.submit(fn=search_free_text_animated, inputs=[txt_input, num_results, lang_choice], outputs=[loader_results, out_results])
 
     btn_find.click(fn=find_book_cards_animated, inputs=[txt_title], outputs=[loader_cards, cards_view, books_state, out_results])
     txt_title.submit(fn=find_book_cards_animated, inputs=[txt_title], outputs=[loader_cards, cards_view, books_state, out_results])
 
-    cards_view.click(
-        fn=on_card_click_animated,
-        inputs=[cards_view, books_state, num_results, lang_choice],
-        outputs=[loader_results, out_results]
-    )
-
-    # Quando clicco su "Ricerca per Trama", pulisco il Tab "Mi è piaciuto"
-    tab_trama.select(
-        fn=clear_find_tab,
-        inputs=None,
-        outputs=[txt_title, cards_view, out_results]
-    )
-
-    # Quando clicco su "Mi è piaciuto", pulisco il Tab "Ricerca per Trama" e i risultati globali
-    tab_libro.select(
-        fn=clear_search_tab,
-        inputs=None,
-        outputs=[txt_input, out_results]
-    )
+    cards_view.click(fn=on_card_click_animated, inputs=[cards_view, books_state, num_results, lang_choice], outputs=[loader_results, out_results])
+
+    tab_trama.select(fn=clear_find_tab, inputs=None, outputs=[txt_title, cards_view, out_results])
+    tab_libro.select(fn=clear_search_tab, inputs=None, outputs=[txt_input, out_results])
 
 print("Avvio Gradio...")
 demo.launch(share=True, debug=True)