test static inputs

agents.py

@@ -55,7 +55,7 @@ model = Gemma3ForConditionalGeneration.from_pretrained(
     # quantization_config=get_quantization_config(),
     device_map="auto",
     torch_dtype=torch.float16,
-)
+).eval()
 
 
 def describe_with_voice_dummy(state: State) -> State:
@@ -76,6 +76,8 @@ def describe_with_voice(state: State) -> State:
     caption = state["caption"]
     voice = state["voice"]
 
+    caption = "A golden retriever that seems to be smiling straight to the camera"
+
     # Voice prompt templates
     voice_prompts = {
         "scurvy-ridden pirate": "You are a scurvy-ridden pirate, angry and drunk.",
@@ -118,6 +120,7 @@ def describe_with_voice(state: State) -> State:
 def caption_image(state: State) -> State:
     # image is PIL
     image = state["image"]
+    image = "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"
 
     # Load models (in practice, do this once and cache)
     messages = [