Update app.py

app.py

@@ -51,7 +51,6 @@ auth_token = os.environ.get("TOKEN_FROM_SECRET") or True
 tokenizer = AutoTokenizer.from_pretrained("vikhyatk/moondream2")
 moondream = AutoModelForCausalLM.from_pretrained(
     "vikhyatk/moondream-next",
-    revision="d197f98731eb17b21708da2f3c69e8d295e683ed",
     trust_remote_code=True,
     torch_dtype=torch.float16,
     device_map={"": "cuda"},
@@ -89,8 +88,7 @@ def answer_question(img, prompt):
         yield buffer.strip(), "Thinking..."
 
     answer = queue.get()
-    # yield answer["answer"], answer["thought"]
-    yield answer["answer"], ""
+    yield answer["answer"], answer["thought"]
 
 
 @spaces.GPU(duration=10)
@@ -264,7 +262,7 @@ css = """
         font-size: 1.4rem !important;
     }
 
-    .chain-of-thought span p {
+    .chain-of-thought {
         opacity: 0.7 !important;
     }
 
@@ -278,12 +276,12 @@ css = """
         opacity: 0.3;
     }
 
-    body, body gradio-app {
-        background: none !important;
-    }
 """
 
 with gr.Blocks(title="moondream vl (new)", css=css, js=js) as demo:
+    if IN_SPACES:
+        gr.HTML("<style>body, body gradio-app { background: none !important; }</style>")
+
     gr.Markdown(
         """
         # 🌔 moondream vl (new)
@@ -343,7 +341,7 @@ with gr.Blocks(title="moondream vl (new)", css=css, js=js) as demo:
                     gr.Markdown("Coming soon!")
 
         with gr.Column():
-            thought = gr.Markdown(elem_classes=["chain-of-thought"])
+            thought = gr.Markdown(elem_classes=["chain-of-thought"], line_breaks=True)
             output = gr.Markdown(label="Response", elem_classes=["output-text"])
             ann = gr.Image(visible=False)