Update app.py

app.py

@@ -8,25 +8,19 @@ from transformers import AutoProcessor, Gemma3ForConditionalGeneration, TextIter
 import spaces
 import time
 
-TITLE = "  מודל מבוסס גמה 3 ליצירת שירים מטופשים בעברית   "
+TITLE = " google/gemma-3-270m-it  "
 DESCRIPTION= """    
-       ניתן לבקש שיר על בסיס טקסט, תמונה ווידאו   
-       
-       בכל פעם, יווצר שיר שונה, אז אם לא אהבתם, אפשר לנסות שוב עם אותו הפרומפט    
-       
-    [המודל זמין להורדה](https://huggingface.co/Norod78/gemma-3_4b_hebrew-lyrics-finetune)    
-
-      המודל כּוּיַּיל ע״י [דורון אדלר](https://linktr.ee/Norod78)
+       It's so small
        """
 
 # model config
-model_4b_name = "Norod78/gemma-3_4b_hebrew-lyrics-finetune"
-model_4b = Gemma3ForConditionalGeneration.from_pretrained(
-    model_4b_name,
+model_270m_name = "google/gemma-3-270m-it"
+model_270m = Gemma3ForConditionalGeneration.from_pretrained(
+    model_270m_name,
     device_map="auto",
     torch_dtype=torch.bfloat16
 ).eval()
-processor_4b = AutoProcessor.from_pretrained(model_4b_name)
+processor_4b = AutoProcessor.from_pretrained(model_270m_name)
 # I will add timestamp later
 def extract_video_frames(video_path, num_frames=8):
     cap = cv2.VideoCapture(video_path)
@@ -90,31 +84,6 @@ def format_conversation_history(chat_history):
 
 @spaces.GPU(duration=120)
 def generate_response(input_data, chat_history, max_new_tokens, system_prompt, temperature, top_p, top_k, repetition_penalty):
-    """
-    Creates silly song lyrics in Hebrew based on user input and conversation history.
-
-    Args:
-        input_data (dict or str): 
-            - If dict: must include 'text' (str) and optional 'files' (list of image/video file paths).
-            - If str: treated as plain text input.
-        chat_history (list of dict): 
-            Sequence of past messages, each with keys 'role' and 'content'.
-        max_new_tokens (int): 
-            Maximum number of tokens to generate for the response.
-        system_prompt (str): 
-            Optional system-level instruction to guide the style and content of the response.
-        temperature (float): 
-            Sampling temperature; higher values yield more diverse outputs.
-        top_p (float): 
-            Nucleus sampling threshold for cumulative probability selection.
-        top_k (int): 
-            Limits sampling to the top_k most likely tokens at each step.
-        repetition_penalty (float): 
-            Penalty factor to discourage the model from repeating the same tokens.
-
-    Yields:
-        str: Streaming chunks of the generated Hebrew song lyrics in real time.
-    """
     
     if isinstance(input_data, dict) and "text" in input_data:
         text = input_data["text"]
@@ -132,7 +101,7 @@ def generate_response(input_data, chat_history, max_new_tokens, system_prompt, t
         messages[-1]["content"].extend(new_message["content"])
     else:
         messages.append(new_message)
-    model = model_4b
+    model = model_270m
     processor = processor_4b
     inputs = processor.apply_chat_template(
         messages,
@@ -167,10 +136,10 @@ chat_interface = gr.ChatInterface(
         gr.Slider(label="Max new tokens", minimum=100, maximum=2000, step=1, value=512),
         gr.Textbox(
             label="System Prompt",
-            value="אתה משורר ישראלי, כותב שירים בעברית",
+            value="You are a very helpful multimodal assistant",
             lines=4,
-            placeholder="שנה את ההגדרות של המודל",
-            text_align = 'right', rtl = True
+            placeholder="Change the settings",
+            text_align = 'left', rtl = False
         ),
         gr.Slider(label="Temperature", minimum=0.1, maximum=2.0, step=0.1, value=0.2),
         gr.Slider(label="Top-p", minimum=0.05, maximum=1.0, step=0.05, value=0.4),
@@ -178,20 +147,19 @@ chat_interface = gr.ChatInterface(
         gr.Slider(label="Repetition Penalty", minimum=1.0, maximum=2.0, step=0.05, value=1.1),
     ],
     examples=[
-        [{"text": "כתוב לי בבקשה שיר המתאר את התמונה", "files": ["examples/image1.jpg"]}],
-        [{"text": "תפוח אדמה עם חרדה חברתית"}]
+        [{"text": "Write a poem which describes this image", "files": ["examples/image1.jpg"]}],
     ],
     textbox=gr.MultimodalTextbox(
-        rtl=True,
+        rtl=False,
         label="קלט",
         file_types=["image", "video"],
         file_count="multiple",
-        placeholder="בקשו שיר ו/או העלו תמונה",        
+        placeholder="Input text, image or video",        
     ),
     cache_examples=False,
     type="messages",
     fill_height=True,
-    stop_btn="הפסק",
+    stop_btn="Stop",
     css_paths=["style.css"],
     multimodal=True,
     title=TITLE,