Fix translation interface with correct megalaa models

- Use correct model names: megalaa/coptic-english-translator & megalaa/english-coptic-translator
- Add trust_remote_code=True for custom pipeline code
- Implement virtual Coptic keyboard in Gradio layout
- Add dialect selection (Sahidic/Bohairic) with proper parameters
- Improve UI with better layout and examples

Fixes model loading errors and missing keyboard interface.

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

app.py

@@ -1,214 +1,243 @@
 #!/usr/bin/env python3
 """
 Coptic Translation Interface - Hugging Face Space
-Supports Coptic↔English translation using fine-tuned MEGALAA models
+Supports Coptic↔English translation using megalaa models
 """
 
 import gradio as gr
-import torch
-from transformers import AutoTokenizer, AutoModelForSeq2SeqLM
-
-# Coptic-Greek character mappings for model preprocessing
-COPTIC_TO_GREEK = {
-    "ⲁ": "α", "ⲃ": "β", "ⲅ": "γ", "ⲇ": "δ", "ⲉ": "ε", "ⲋ": "ϛ",
-    "ⲍ": "ζ", "ⲏ": "η", "ⲑ": "θ", "ⲓ": "ι", "ⲕ": "κ", "ⲗ": "λ",
-    "ⲙ": "μ", "ⲛ": "ν", "ⲝ": "ξ", "ⲟ": "ο", "ⲡ": "π", "ⲣ": "ρ",
-    "ⲥ": "σ", "ⲧ": "τ", "ⲩ": "υ", "ⲫ": "φ", "ⲭ": "χ", "ⲯ": "ψ", "ⲱ": "ω",
-    "ϣ": "ʃ", "ϥ": "f", "ϧ": "x", "ϩ": "h", "ϫ": "ɟ", "ϭ": "c", "ϯ": "ti",
-    "Ⲁ": "Α", "Ⲃ": "Β", "Ⲅ": "Γ", "Ⲇ": "Δ", "Ⲉ": "Ε", "Ⲍ": "Ζ", "Ⲏ": "Η",
-    "Ⲑ": "Θ", "Ⲓ": "Ι", "Ⲕ": "Κ", "Ⲗ": "Λ", "Ⲙ": "Μ", "Ⲛ": "Ν", "Ⲝ": "Ξ",
-    "Ⲟ": "Ο", "Ⲡ": "Π", "Ⲣ": "Ρ", "Ⲥ": "Σ", "Ⲧ": "Τ", "Ⲩ": "Υ", "Ⲫ": "Φ",
-    "Ⲭ": "Χ", "Ⲯ": "Ψ", "Ⲱ": "Ω", "Ϣ": "Ʃ", "Ϥ": "F", "Ϧ": "X", "Ϩ": "H",
-    "Ϫ": "Ɉ", "Ϭ": "C", "Ϯ": "TI"
-}
-
-GREEK_TO_COPTIC = {v: k for k, v in COPTIC_TO_GREEK.items()}
-
-def greekify(coptic_text):
-    """Convert Coptic Unicode to Greek transcription"""
-    return "".join(COPTIC_TO_GREEK.get(c.lower(), c.lower()) for c in coptic_text)
-
-def degreekify(greek_text):
-    """Convert Greek transcription back to Coptic Unicode"""
-    result = []
-    i = 0
-    while i < len(greek_text):
-        if i < len(greek_text) - 1 and greek_text[i:i+2].lower() == 'ti':
-            result.append(GREEK_TO_COPTIC.get('ti', greek_text[i:i+2]))
-            i += 2
-        else:
-            result.append(GREEK_TO_COPTIC.get(greek_text[i], greek_text[i]))
-            i += 1
-    return ''.join(result)
+from transformers import pipeline
+
+# Coptic alphabet for virtual keyboard
+COPTIC_LETTERS = [
+    'ⲁ', 'ⲃ', 'ⲅ', 'ⲇ', 'ⲉ', 'ⲍ', 'ⲏ', 'ⲑ', 'ⲓ', 'ⲕ', 'ⲗ', 'ⲙ',
+    'ⲛ', 'ⲝ', 'ⲟ', 'ⲡ', 'ⲣ', 'ⲥ', 'ⲧ', 'ⲩ', 'ⲫ', 'ⲭ', 'ⲯ', 'ⲱ',
+    'ϣ', 'ϥ', 'ϧ', 'ϩ', 'ϫ', 'ϭ', 'ϯ'
+]
 
-# Model loading with caching
-coptic_to_english_model = None
-english_to_coptic_model = None
-device = "cuda" if torch.cuda.is_available() else "cpu"
+# Model caching
+coptic_to_english_pipe = None
+english_to_coptic_pipe = None
 
 def load_coptic_to_english():
-    global coptic_to_english_model
-    if coptic_to_english_model is None:
-        model_name = "Norelad/coptic-megalaa-finetuned"
-        tokenizer = AutoTokenizer.from_pretrained(model_name)
-        model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)
-        coptic_to_english_model = (tokenizer, model)
-    return coptic_to_english_model
+    """Load Coptic → English translation pipeline"""
+    global coptic_to_english_pipe
+    if coptic_to_english_pipe is None:
+        coptic_to_english_pipe = pipeline(
+            model="megalaa/coptic-english-translator",
+            trust_remote_code=True
+        )
+    return coptic_to_english_pipe
 
 def load_english_to_coptic():
-    global english_to_coptic_model
-    if english_to_coptic_model is None:
-        model_name = "megalaa/english-coptic-translator"
-        tokenizer = AutoTokenizer.from_pretrained(model_name)
-        model = AutoModelForSeq2SeqLM.from_pretrained(model_name).to(device)
-        english_to_coptic_model = (tokenizer, model)
-    return english_to_coptic_model
-
-def translate_coptic_to_english(text, dialect='cop-sa'):
+    """Load English → Coptic translation pipeline"""
+    global english_to_coptic_pipe
+    if english_to_coptic_pipe is None:
+        english_to_coptic_pipe = pipeline(
+            model="megalaa/english-coptic-translator",
+            trust_remote_code=True
+        )
+    return english_to_coptic_pipe
+
+def translate_coptic_to_english(text, dialect):
     """Translate Coptic to English"""
+    if not text or not text.strip():
+        return "Please enter Coptic text to translate."
+
     try:
-        tokenizer, model = load_coptic_to_english()
-
-        # Dialect tags
-        DIALECT_TAGS = {'cop-sa': 'з', 'cop-bo': 'б', 'cop': 'з'}
-        dialect_tag = DIALECT_TAGS.get(dialect, 'з')
-
-        # Preprocess: Convert to Greek transcription and add dialect tag
-        greek_input = greekify(text.lower())
-        greek_input = f"{dialect_tag} {greek_input}"
-
-        # Generate translation
-        inputs = tokenizer(greek_input, return_tensors="pt", padding=True).to(device)
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=128,
-            num_beams=5,
-            early_stopping=True
-        )
+        pipe = load_coptic_to_english()
+
+        # Use from_bohairic parameter if Bohairic dialect selected
+        if dialect == "Bohairic":
+            result = pipe(text, from_bohairic=True)
+        else:
+            result = pipe(text)
 
-        return tokenizer.decode(outputs[0], skip_special_tokens=True)
+        return result['translation']
     except Exception as e:
-        return f"Translation error: {e}"
+        return f"Translation error: {str(e)}"
 
-def translate_english_to_coptic(text):
+def translate_english_to_coptic(text, dialect):
     """Translate English to Coptic"""
+    if not text or not text.strip():
+        return "Please enter English text to translate."
+
     try:
-        tokenizer, model = load_english_to_coptic()
-
-        # Generate translation
-        inputs = tokenizer(text, return_tensors="pt", padding=True).to(device)
-        outputs = model.generate(
-            **inputs,
-            max_new_tokens=128,
-            num_beams=5,
-            early_stopping=True
-        )
+        pipe = load_english_to_coptic()
+
+        # Use to_bohairic parameter if Bohairic dialect selected
+        if dialect == "Bohairic":
+            result = pipe(text, to_bohairic=True)
+        else:
+            result = pipe(text)
 
-        # Convert Greek output to Coptic
-        greek_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
-        return degreekify(greek_output)
+        return result['translation']
     except Exception as e:
-        return f"Translation error: {e}"
+        return f"Translation error: {str(e)}"
+
+def add_letter(current_text, letter):
+    """Add a Coptic letter to the current text"""
+    return current_text + letter if current_text else letter
+
+def add_space(current_text):
+    """Add a space to the current text"""
+    return current_text + " " if current_text else " "
+
+def backspace(current_text):
+    """Remove last character from current text"""
+    return current_text[:-1] if current_text else ""
+
+def clear_text():
+    """Clear all text"""
+    return ""
 
 # Example texts
 COPTIC_EXAMPLES = [
-    ("ϯⲛⲁⲃⲱⲕ ⲉⲡⲏⲓ", "I will go to the house"),
-    ("ⲡⲉⲭⲣⲓⲥⲧⲟⲥ ⲡⲉ ⲡⲛⲟⲩⲧⲉ", "Christ is God"),
-    ("ⲁⲓⲛⲁⲩ ⲉⲡⲣⲱⲙⲉ", "I saw the man"),
+    ["ϯⲛⲁⲃⲱⲕ ⲉⲡⲏⲓ", "Sahidic"],
+    ["ⲡⲉⲭⲣⲓⲥⲧⲟⲥ ⲡⲉ ⲡⲛⲟⲩⲧⲉ", "Sahidic"],
+    ["ⲁⲓⲛⲁⲩ ⲉⲡⲣⲱⲙⲉ", "Sahidic"],
 ]
 
 ENGLISH_EXAMPLES = [
-    "The Lord is good",
-    "I am a teacher",
-    "We give thanks to God",
+    ["The Lord is good", "Sahidic"],
+    ["I am a teacher", "Sahidic"],
+    ["We give thanks to God", "Sahidic"],
 ]
 
-# Gradio Interface
+# Create Gradio Interface
 with gr.Blocks(title="Coptic Translation Interface", theme=gr.themes.Soft()) as demo:
     gr.Markdown("""
     # 🔮 Coptic Translation Interface
 
-    Translate between Coptic and English using fine-tuned MEGALAA models:
-    - **Coptic → English**: `Norelad/coptic-megalaa-finetuned`
+    Translate between Coptic and English using specialized models from [megalaa](https://huggingface.co/megalaa):
+    - **Coptic → English**: `megalaa/coptic-english-translator`
     - **English → Coptic**: `megalaa/english-coptic-translator`
 
-    Based on 50,000+ parallel sentences from CopticScriptorium corpus.
+    Based on neural machine translation models trained on Coptic-English parallel corpus.
     """)
 
-    with gr.Tab("Coptic → English"):
-        with gr.Row():
-            with gr.Column():
-                cop_input = gr.Textbox(
-                    label="Coptic Text",
-                    placeholder="Enter Coptic text (Unicode)...",
-                    lines=5
-                )
-                cop_dialect = gr.Radio(
-                    choices=[("Sahidic", "cop-sa"), ("Bohairic", "cop-bo")],
-                    value="cop-sa",
-                    label="Dialect"
-                )
-                cop_translate_btn = gr.Button("Translate to English", variant="primary")
-
-            with gr.Column():
-                cop_output = gr.Textbox(
-                    label="English Translation",
-                    lines=5,
-                    interactive=False
-                )
-
-        gr.Examples(
-            examples=[[ex[0], "cop-sa"] for ex in COPTIC_EXAMPLES],
-            inputs=[cop_input, cop_dialect],
-            outputs=cop_output,
-            fn=translate_coptic_to_english,
-            label="Example Coptic Texts"
-        )
-
-        cop_translate_btn.click(
-            fn=translate_coptic_to_english,
-            inputs=[cop_input, cop_dialect],
-            outputs=cop_output
-        )
-
-    with gr.Tab("English → Coptic"):
-        with gr.Row():
-            with gr.Column():
-                eng_input = gr.Textbox(
-                    label="English Text",
-                    placeholder="Enter English text...",
-                    lines=5
-                )
-                eng_translate_btn = gr.Button("Translate to Coptic", variant="primary")
-
-            with gr.Column():
-                eng_output = gr.Textbox(
-                    label="Coptic Translation",
-                    lines=5,
-                    interactive=False
-                )
-
-        gr.Examples(
-            examples=[[ex] for ex in ENGLISH_EXAMPLES],
-            inputs=eng_input,
-            outputs=eng_output,
-            fn=translate_english_to_coptic,
-            label="Example English Texts"
-        )
-
-        eng_translate_btn.click(
-            fn=translate_english_to_coptic,
-            inputs=eng_input,
-            outputs=eng_output
-        )
+    with gr.Tabs():
+        # Tab 1: Coptic → English
+        with gr.TabItem("Coptic → English"):
+            gr.Markdown("### Translate Coptic text to English")
+
+            with gr.Row():
+                with gr.Column(scale=1):
+                    cop_input = gr.Textbox(
+                        label="Coptic Text",
+                        placeholder="Enter Coptic text or use the virtual keyboard below...",
+                        lines=8,
+                        max_lines=15
+                    )
+
+                    cop_dialect = gr.Radio(
+                        choices=["Sahidic", "Bohairic"],
+                        value="Sahidic",
+                        label="Coptic Dialect"
+                    )
+
+                    # Virtual Coptic Keyboard
+                    with gr.Group():
+                        gr.Markdown("**Virtual Coptic Keyboard**")
+
+                        # Create keyboard in rows of 8
+                        for i in range(0, len(COPTIC_LETTERS), 8):
+                            with gr.Row():
+                                for letter in COPTIC_LETTERS[i:i+8]:
+                                    btn = gr.Button(letter, size="sm", scale=1)
+                                    btn.click(
+                                        fn=lambda current, l=letter: add_letter(current, l),
+                                        inputs=[cop_input],
+                                        outputs=[cop_input]
+                                    )
+
+                        with gr.Row():
+                            space_btn = gr.Button("Space", size="sm", scale=2)
+                            back_btn = gr.Button("⌫ Backspace", size="sm", scale=2)
+                            clear_btn = gr.Button("Clear", size="sm", scale=1)
+
+                        space_btn.click(fn=add_space, inputs=[cop_input], outputs=[cop_input])
+                        back_btn.click(fn=backspace, inputs=[cop_input], outputs=[cop_input])
+                        clear_btn.click(fn=clear_text, outputs=[cop_input])
+
+                    cop_translate_btn = gr.Button("🔄 Translate to English", variant="primary", size="lg")
+
+                with gr.Column(scale=1):
+                    cop_output = gr.Textbox(
+                        label="English Translation",
+                        lines=8,
+                        max_lines=15,
+                        interactive=False
+                    )
+
+            gr.Examples(
+                examples=COPTIC_EXAMPLES,
+                inputs=[cop_input, cop_dialect],
+                outputs=cop_output,
+                fn=translate_coptic_to_english,
+                cache_examples=False,
+                label="📖 Example Coptic Texts"
+            )
+
+            cop_translate_btn.click(
+                fn=translate_coptic_to_english,
+                inputs=[cop_input, cop_dialect],
+                outputs=cop_output
+            )
+
+        # Tab 2: English → Coptic
+        with gr.TabItem("English → Coptic"):
+            gr.Markdown("### Translate English text to Coptic")
+
+            with gr.Row():
+                with gr.Column(scale=1):
+                    eng_input = gr.Textbox(
+                        label="English Text",
+                        placeholder="Enter English text...",
+                        lines=8,
+                        max_lines=15
+                    )
+
+                    eng_dialect = gr.Radio(
+                        choices=["Sahidic", "Bohairic"],
+                        value="Sahidic",
+                        label="Target Coptic Dialect"
+                    )
+
+                    eng_translate_btn = gr.Button("🔄 Translate to Coptic", variant="primary", size="lg")
+
+                with gr.Column(scale=1):
+                    eng_output = gr.Textbox(
+                        label="Coptic Translation",
+                        lines=8,
+                        max_lines=15,
+                        interactive=False
+                    )
+
+            gr.Examples(
+                examples=ENGLISH_EXAMPLES,
+                inputs=[eng_input, eng_dialect],
+                outputs=eng_output,
+                fn=translate_english_to_coptic,
+                cache_examples=False,
+                label="📖 Example English Texts"
+            )
+
+            eng_translate_btn.click(
+                fn=translate_english_to_coptic,
+                inputs=[eng_input, eng_dialect],
+                outputs=eng_output
+            )
 
     gr.Markdown("""
     ---
     ### About
-    This interface uses fine-tuned MarianMT models trained on the CopticScriptorium parallel corpus.
-    The models support bidirectional translation between Sahidic/Bohairic Coptic and English.
 
-    **Note**: For best results with Coptic input, use proper Unicode Coptic characters (U+2C80–U+2CFF).
+    This interface uses neural machine translation models trained on Coptic-English parallel corpus:
+    - **Models**: [megalaa/coptic-english-translator](https://huggingface.co/megalaa/coptic-english-translator) & [megalaa/english-coptic-translator](https://huggingface.co/megalaa/english-coptic-translator)
+    - **Dialects**: Supports both Sahidic (default) and Bohairic Coptic
+    - **Input**: Use proper Unicode Coptic characters (U+2C80–U+2CFF) or the virtual keyboard
+
+    **Research**: Based on work by Enis & Megalaa (2024) - "Ancient voices, modern technology: Low-resource neural machine translation for coptic texts"
     """)
 
 if __name__ == "__main__":