auto detection of language input

app/app.py

@@ -25,6 +25,7 @@ from transformers import pipeline, AutoTokenizer, AutoModelForCausalLM
 import threading
 from collections import defaultdict
 from datasets import load_dataset
+from lang_model_router import detect_language_code, get_language_name_and_model
 
 
 BASE_MODEL = os.getenv("MODEL", "google/gemma-3-12b-pt")
@@ -396,12 +397,29 @@ def respond(
     language: str,
     temperature: Optional[float] = None,
     seed: Optional[int] = None,
+    auto_detect: bool = True,
 ) -> list:
-    """Respond to the user message with a system message
-
-    Return the history with the new message"""
-    messages = format_history_as_messages(history)
-
+    """Respond to the user message with system prompt in auto-detected or selected language."""
+    # Get last user message
+    user_input = ""
+    for msg in reversed(history):
+        if msg["role"] == "user":
+            user_input = msg["content"]
+            break
+
+    # Determine language
+    if auto_detect:
+        lang_code = detect_language_code(user_input)
+        language, _ = get_language_name_and_model(lang_code)
+
+    # Load system prompt
+    system_prompt = LANGUAGES.get(language, LANGUAGES["English"])
+
+    # Format message list with system prompt prepended
+    messages = [{"role": "system", "content": system_prompt}]
+    messages.extend(format_history_as_messages(history))
+
+    # Generate response
     if ZERO_GPU:
         content = call_pipeline(messages)
     else:
@@ -416,6 +434,7 @@ def respond(
         )
         content = response.choices[0].message.content
 
+    # Add response to history
     message = gr.ChatMessage(role="assistant", content=content)
     history.append(message)
     return history

app/lang_model_router.py

@@ -0,0 +1,35 @@
+
+from langdetect import detect, DetectorFactory
+from transformers import AutoTokenizer, AutoModelForCausalLM
+import os
+DetectorFactory.seed = 0 
+LANGUAGE_MAP = {
+    "en": {"name": "English", "model": "openai-community/gpt2"},
+    "fr": {"name": "French", "model": "dbddv01/gpt2-french-small"},
+    "es": {"name": "Spanish", "model": "datificate/gpt2-small-spanish"},
+    "de": {"name": "German", "model": "deepset/gbert-base"},
+    "hi": {"name": "Hindi", "model": "ai4bharat/indic-bert"},
+    "mr": {"name": "Marathi", "model": "ai4bharat/indic-bert"},
+    "ja": {"name": "Japanese", "model": "rinna/japanese-gpt2-medium"},
+    "zh-cn": {"name": "Chinese", "model": "uer/gpt2-chinese-cluecorpusswwm"},
+    "ru": {"name": "Russian", "model": "sberbank-ai/rugpt3small_based_on_gpt2"},
+    "pt": {"name": "Portuguese", "model": "pierreguillou/gpt2-small-portuguese"},
+    "it": {"name": "Italian", "model": "dbddv01/gpt2-italian"},
+    "nl": {"name": "Dutch", "model": "GroNLP/gpt2-small-dutch"}
+}
+
+
+def detect_language_code(text: str) -> str:
+    try:
+        return detect(text)
+    except Exception:
+        return "en"  # fallback
+
+def get_language_name_and_model(lang_code: str) -> tuple[str, str]:
+    return LANGUAGE_MAP.get(lang_code, LANGUAGE_MAP["en"])
+
+def get_model_by_name(language_name: str) -> str:
+    for code, (name, model) in LANGUAGE_MAP.items():
+        if name.lower() == language_name.lower():
+            return model
+    return LANGUAGE_MODEL_MAP["en"][1]