Final model upload - BLEU: 0.2490

README.md

@@ -11,18 +11,31 @@ tags:
 
 # English-Japanese Transformer
 
-Model Transformer untuk terjemahan English → Japanese, dilatih dengan Keras 3.
+A Transformer model trained for English to Japanese machine translation, built using Keras 3.
 
 ## Performance
+The model was trained for 24 epochs, stopping early to prevent overfitting.
+
 - **Validation Accuracy**: 0.9088
 - **Average Character BLEU Score**: 0.2490
 
 ## Usage (Loading Safely)
-Model ini harus dimuat menggunakan mekanisme Keras:
+This model is stored in the secure Keras v3 format (`.keras`) and should be loaded as follows:
 
 ```python
 import keras
-# Ini adalah cara yang aman untuk memuat bobot dari file .keras
+from keras import ops
+
+# Load the full model (including architecture and weights)
 model = keras.models.load_model("transformer_model.keras") 
-# ... Lanjutkan dengan proses inferensi Anda
+
+# IMPORTANT: You must extract the individual layers/sub-models for step-by-step inference.
+# Example of manual inference call (similar to Cell 3):
+# 1. Get layers:
+# embedding_layer = model.get_layer('token_and_position_embedding')
+# transformer_encoder_layer = model.get_layer('transformer_encoder')
+# decoder_model = model.get_layer('decoder') 
+
+# 2. Run inference loop with training=False
+# ...
 ```

transformer_model.keras

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:bb91b23cb164fc6979f66c319627adfa1c342f98ad0bc9f170f947be0f1339d2
+oid sha256:aa8625619232535a81db1089ce457052ec28dc4b38c8fafa5b69f39806dbbc58
 size 95224033