Update app.py

app.py

@@ -31,3 +31,69 @@ if is_allowed_time():
         st.audio(audio_output, format='audio/mp3')
 else:
     st.write("Translation service is available after 6 PM IST.")
+
+
+import sacrebleu
+import torch
+def evaluate_bleu_score(model, tokenizer, test_data):
+    references = []
+    hypotheses = []
+
+    # Ensure the model is on the correct device
+    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+    model.to(device)
+
+    for item in test_data:
+        english_text = item["english"]
+        expected_hindi = item["hindi"]
+
+        # Tokenize and move inputs to the correct device
+        inputs = tokenizer(english_text, return_tensors="pt", padding=True, truncation=True).to(device)
+        translated_tokens = model.generate(**inputs)
+        translated_hindi = tokenizer.decode(translated_tokens[0], skip_special_tokens=True)
+
+        references.append([expected_hindi])
+        hypotheses.append(translated_hindi)
+
+    # Calculate BLEU score
+    bleu = sacrebleu.corpus_bleu(hypotheses, references)
+    return bleu.score
+
+test_data = [
+    {"english": "Hello, how are you?", "hindi": "नमस्ते, आप कैसे हैं?"},
+    {"english": "What is your name?", "hindi": "आपका नाम क्या है?"},
+    {"english": "I love to play football.", "hindi": "मुझे फुटबॉल खेलना बहुत पसंद है।"},
+    {"english": "How old are you?", "hindi": "आपकी उम्र क्या है?"},
+    {"english": "I am from India.", "hindi": "मैं भारत से हूँ।"},
+    {"english": "What do you do?", "hindi": "आप क्या करते हैं?"},
+    {"english": "I like to read books.", "hindi": "मुझे किताबें पढ़ना पसंद है।"},
+    {"english": "Do you speak English?", "hindi": "क्या आप अंग्रेज़ी बोलते हैं?"},
+    {"english": "I have a dog.", "hindi": "मेरे पास एक कुत्ता है।"},
+    {"english": "What is your favorite food?", "hindi": "आपका पसंदीदा खाना क्या है?"},
+    {"english": "I love to travel.", "hindi": "मुझे यात्रा करना बहुत पसंद है।"},
+    {"english": "How much does it cost?", "hindi": "इसकी कीमत क्या है?"},
+    {"english": "I am happy to meet you.", "hindi": "आपसे मिलने में मुझे खुशी है।"}
+    
+]
+training_args = Seq2SeqTrainingArguments(
+    output_dir="./results",
+    evaluation_strategy="epoch",
+    learning_rate=2e-5,
+    per_device_train_batch_size=16,
+    per_device_eval_batch_size=16,
+    weight_decay=0.01,
+    save_total_limit=3,
+    num_train_epochs=10,  # Increase number of epochs
+    predict_with_generate=True,
+)
+
+trainer = Seq2SeqTrainer(
+    model=model,
+    args=training_args,
+    train_dataset=tokenized_datasets,
+    eval_dataset=tokenized_datasets,
+)
+
+trainer.train()
+bleu_score = evaluate_bleu_score(model, tokenizer, test_data)
+print(f"Accuracy: {bleu_score:.2f}%")