Actual text streaming

Also yield `...` when Stream starts, as a UX feedback for "forward-pass started"
Live on this replicated Space: https://huggingface.co/spaces/cbensimon/SmolVLM

app.py

@@ -76,8 +76,8 @@ def model_inference(
 
     thread = Thread(target=model.generate, kwargs=generation_args)
     thread.start()
-    thread.join()
-    
+
+    yield "..."
     buffer = ""