Update app.py

app.py

@@ -30,7 +30,7 @@ Question: {question}
 
 To answer this question:
 1. If you need current information or facts you're unsure about, use the search tool
-2. If you need to transcribe audio, import Transcriber from smolagents and instantiate 'transcriber = Transcriber()'
+2. If you need to transcribe audio, import smolagents and instantiate 'transcriber = smolagents.Transcriber()'
 3. Write Python code to solve the problem
 4. Make sure your final answer is clear and direct
 
@@ -69,7 +69,7 @@ class BasicAgent:
                 # Initialize the model
                 print("🤖 Initializing TransformersModel...")
                 self.model = TransformersModel(
-                    model_id="google/gemma-3-4b-it",
+                    model_id="google/gemma-3-12b-it",
                     torch_dtype=torch.bfloat16,
                     device_map="auto",
                 )
@@ -81,7 +81,7 @@ class BasicAgent:
                     if self.model.tokenizer.pad_token is None:
                         self.model.tokenizer.pad_token = self.model.tokenizer.eos_token
                     # Set default padding behavior for FlashAttention
-                    self.model.tokenizer.pad_to_multiple_of = 8
+                    self.model.tokenizer.pad_to_multiple_of = 64
                     print("✅ Applied tokenizer padding fix for FlashAttention alignment")
                 
                 # If the model has a processor with tokenizer, fix that too
@@ -89,7 +89,7 @@ class BasicAgent:
                     self.model.processor.tokenizer.padding_side = "left"
                     if self.model.processor.tokenizer.pad_token is None:
                         self.model.processor.tokenizer.pad_token = self.model.processor.tokenizer.eos_token
-                    self.model.processor.tokenizer.pad_to_multiple_of = 8
+                    self.model.processor.tokenizer.pad_to_multiple_of = 64
                     print("✅ Applied processor tokenizer padding fix")