Simplify to single Shako v4 model only - remove model selection and use single model

app.py

@@ -9,27 +9,19 @@ import spaces
 import time
 import os
 
-# model config
-model_12b_name = "anaspro/Shako-4B-it-v4"
-model_4b_name = "anaspro/Shako-4B-it"
+# model config - Single model: Shako v4
+model_name = "anaspro/Shako-4B-it-v4"
 
 # Load token from environment if available
 hf_token = os.getenv("HF_TOKEN")
 
-model_12b = Gemma3ForConditionalGeneration.from_pretrained(
-    model_12b_name,
+model = Gemma3ForConditionalGeneration.from_pretrained(
+    model_name,
     device_map="auto",
     torch_dtype=torch.bfloat16,
     token=hf_token
 ).eval()
-processor_12b = AutoProcessor.from_pretrained(model_12b_name, token=hf_token)
-
-model_4b = Gemma3ForConditionalGeneration.from_pretrained(
-    model_4b_name,
-    device_map="auto",
-    torch_dtype=torch.bfloat16
-).eval()
-processor_4b = AutoProcessor.from_pretrained(model_4b_name)
+processor = AutoProcessor.from_pretrained(model_name, token=hf_token)
 # I will add timestamp later
 def extract_video_frames(video_path, num_frames=8):
     cap = cv2.VideoCapture(video_path)
@@ -92,7 +84,7 @@ def format_conversation_history(chat_history):
     return messages
 
 @spaces.GPU(duration=120)
-def generate_response(input_data, chat_history, model_choice, max_new_tokens, system_prompt, temperature, top_p, top_k, repetition_penalty):
+def generate_response(input_data, chat_history, max_new_tokens, system_prompt, temperature, top_p, top_k, repetition_penalty):
     if isinstance(input_data, dict) and "text" in input_data:
         text = input_data["text"]
         files = input_data.get("files", [])
@@ -109,12 +101,7 @@ def generate_response(input_data, chat_history, model_choice, max_new_tokens, sy
         messages[-1]["content"].extend(new_message["content"])
     else:
         messages.append(new_message)
-    if model_choice == "Gemma 3 12B":
-        model = model_12b
-        processor = processor_12b
-    else:
-        model = model_4b
-        processor = processor_4b
+    # Use the single Shako v4 model
     inputs = processor.apply_chat_template(
         messages,
         add_generation_prompt=True,
@@ -144,11 +131,6 @@ def generate_response(input_data, chat_history, model_choice, max_new_tokens, sy
 demo = gr.ChatInterface(
     fn=generate_response,
     additional_inputs=[
-        gr.Dropdown(
-            label="Model",
-            choices=["Gemma 3 12B", "Gemma 3 4B"],
-            value="Gemma 3 12B"
-        ),
         gr.Slider(label="Max new tokens", minimum=100, maximum=2000, step=1, value=512),
         gr.Textbox(
             label="System Prompt",
@@ -167,8 +149,8 @@ demo = gr.ChatInterface(
     cache_examples=False,
     type="messages",
     description="""
-    # Gemma 3
-    You can pick your model 12B or 4B, upload images or videos, and adjust settings below to customize your experience.
+    # شكو - Shako Iraqi AI
+    نموذج ذكاء عراقي متقدم يتحدث بالعراقي، يدعم الصور والفيديوهات والمحادثات الصوتية.
     """,
     fill_height=True,
     textbox=gr.MultimodalTextbox(