diff --git a/libs/python/agent/agent/adapters/models/__init__.py b/libs/python/agent/agent/adapters/models/__init__.py
index 6811c142..8a5fb00b 100644
--- a/libs/python/agent/agent/adapters/models/__init__.py
+++ b/libs/python/agent/agent/adapters/models/__init__.py
@@ -22,7 +22,7 @@ def load_model(model_name: str, device: str = "auto"):
         )
     cfg = AutoConfig.from_pretrained(model_name, trust_remote_code=True)
     cls = cfg.__class__.__name__
-    print(f"cls: {cls}")
+    # print(f"cls: {cls}")
     if "OpenCUA" in cls:
         return OpenCUAModel(model_name=model_name, device=device)
     return GenericHFModel(model_name=model_name, device=device)
diff --git a/libs/python/agent/agent/adapters/models/opencua.py b/libs/python/agent/agent/adapters/models/opencua.py
index f24dfa6b..f8abf4a6 100644
--- a/libs/python/agent/agent/adapters/models/opencua.py
+++ b/libs/python/agent/agent/adapters/models/opencua.py
@@ -37,6 +37,7 @@ class OpenCUAModel:
             torch_dtype="auto",
             device_map=self.device,
             trust_remote_code=True,
+            attn_implementation="sdpa",
         )
         self.image_processor = AutoImageProcessor.from_pretrained(
             self.model_name, trust_remote_code=True
diff --git a/libs/python/agent/agent/loops/opencua.py b/libs/python/agent/agent/loops/opencua.py
index c13875b2..a494377b 100644
--- a/libs/python/agent/agent/loops/opencua.py
+++ b/libs/python/agent/agent/loops/opencua.py
@@ -97,7 +97,7 @@ class OpenCUAConfig(AsyncAgentConfig):
                 },
                 {
                     "type": "text",
-                    "text": instruction
+                    "text": f"Click on {instruction}"
                 }
             ]
         }
@@ -116,8 +116,7 @@ class OpenCUAConfig(AsyncAgentConfig):
         
         # Extract response text
         output_text = response.choices[0].message.content
-        
-        print(output_text)
+        # print(output_text)
         
         # Extract coordinates from pyautogui format
         coordinates = extract_coordinates_from_pyautogui(output_text)
diff --git a/libs/python/agent/pyproject.toml b/libs/python/agent/pyproject.toml
index 811c3a9c..0d382fdf 100644
--- a/libs/python/agent/pyproject.toml
+++ b/libs/python/agent/pyproject.toml
@@ -50,7 +50,7 @@ glm45v-hf = [
 opencua-hf = [
     "accelerate",
     "torch",
-    "transformers>=4.54.0",
+    "transformers==4.53.0",
     "tiktoken>=0.11.0",
     "blobfile>=3.0.0"
 ]
@@ -75,7 +75,7 @@ all = [
     "transformers>=4.54.0",
     # opencua requirements
     "tiktoken>=0.11.0",
-    "blobfile>=3.0.0"
+    "blobfile>=3.0.0",
     # ui requirements
     "gradio>=5.23.3",
     "python-dotenv>=1.0.1",