Update app.py

app.py

@@ -46,7 +46,6 @@ moondream = AutoModelForCausalLM.from_pretrained(
     device_map={"": "cuda"},
     revision=REVISION
 )
-moondream.use_flex_decoding = False
 moondream.eval()
 
 
@@ -116,6 +115,8 @@ def convert_to_entities(text, coords):
 
 @spaces.GPU(duration=30)
 def answer_question(img, prompt, reasoning):
+    moondream.use_flex_decoding = False
+
     buffer = ""
     resp = moondream.query(img, prompt, stream=True, reasoning=reasoning)
     reasoning_text = resp["reasoning"]["text"] if reasoning else "[reasoning disabled]"
@@ -130,6 +131,8 @@ def answer_question(img, prompt, reasoning):
 
 @spaces.GPU(duration=10)
 def caption(img, mode):
+    moondream.use_flex_decoding = False
+
     if img is None:
         yield ""
         return
@@ -147,6 +150,8 @@ def caption(img, mode):
 
 @spaces.GPU(duration=10)
 def detect(img, object, eos_bias):
+    moondream.use_flex_decoding = False
+
     if img is None:
         yield "", gr.update(visible=False, value=None)
         return
@@ -175,6 +180,8 @@ def detect(img, object, eos_bias):
 
 @spaces.GPU(duration=10)
 def point(img, object):
+    moondream.use_flex_decoding = False
+
     if img is None:
         yield "", gr.update(visible=False, value=None)
         return
@@ -200,6 +207,8 @@ def point(img, object):
 
 @spaces.GPU(duration=10)
 def localized_query(img, x, y, question):
+    moondream.use_flex_decoding = False
+
     if img is None:
         yield "", {"text": "", "entities": []}, gr.update(visible=False, value=None)
         return