Update app.py

app.py

@@ -8,11 +8,11 @@ import av
 import gradio as gr
 import spaces
 import torch
-from gradio.utils import get_upload_folder
+from fastrtc import AdditionalOutputs, ReplyOnPause, WebRTC, WebRTCData, get_hf_turn_credentials
 from gradio.processing_utils import save_audio_to_cache
+from gradio.utils import get_upload_folder
 from transformers import AutoModelForImageTextToText, AutoProcessor
 from transformers.generation.streamers import TextIteratorStreamer
-from fastrtc import ReplyOnPause, WebRTCData, WebRTC, AdditionalOutputs, get_hf_turn_credentials
 
 model_id = "google/gemma-3n-E4B-it"
 
@@ -202,12 +202,19 @@ def _generate(message: dict, history: list[dict], system_prompt: str = "", max_n
 
 @spaces.GPU(time_limit=120)
 def generate(data: WebRTCData, history: list[dict], system_prompt: str = "", max_new_tokens: int = 512, image=None):
-    message = {"text": data.textbox, "files": [save_audio_to_cache(data.audio[1], data.audio[0], format="mp3", cache_dir=get_upload_folder())]}
+    files = []
+    if data.audio is not None and data.audio[1].size > 0:
+        files.append(save_audio_to_cache(data.audio[1], data.audio[0], format="mp3", cache_dir=get_upload_folder()))
+    if image is None:
+        files.append(image)
+    message = {
+        "text": data.textbox,
+        "files": [],
+    }
     new_message = {"role": "assistant", "content": ""}
     for output in _generate(message, history, system_prompt, max_new_tokens):
         new_message["content"] += output
         yield AdditionalOutputs(history + [new_message])
-    
 
 
 with gr.Blocks() as demo:
@@ -217,12 +224,12 @@ with gr.Blocks() as demo:
         mode="send",
         variant="textbox",
         rtc_configuration=get_hf_turn_credentials,
-        server_rtc_configuration=get_hf_turn_credentials(ttl=3_600 * 24 * 30)
+        server_rtc_configuration=get_hf_turn_credentials(ttl=3_600 * 24 * 30),
     )
     with gr.Accordion(label="Additional Inputs"):
         sp = gr.Textbox(label="System Prompt", value="You are a helpful assistant.")
         slider = gr.Slider(label="Max New Tokens", minimum=100, maximum=2000, step=10, value=700)
-        image = gr.Image()
+        image = gr.Image(type="filepath")
 
     webrtc.stream(
         ReplyOnPause(generate),  # type: ignore
@@ -230,9 +237,7 @@ with gr.Blocks() as demo:
         outputs=[chatbot],
         concurrency_limit=100,
     )
-    webrtc.on_additional_outputs(
-        lambda old, new: new, inputs=[chatbot], outputs=[chatbot], concurrency_limit=100
-    )
+    webrtc.on_additional_outputs(lambda old, new: new, inputs=[chatbot], outputs=[chatbot], concurrency_limit=100)
 
 if __name__ == "__main__":
     demo.launch()