feat: 支持输入url与text

app.py

@@ -13,18 +13,23 @@ model = Qwen2VLForConditionalGeneration.from_pretrained("Qwen/Qwen2-VL-7B-Instru
 processor = AutoProcessor.from_pretrained("Qwen/Qwen2-VL-7B-Instruct")
 
 @spaces.GPU
-def infer(n):
-    if len(n) < 1:
-        n = "请将图里文字转成markdown"
+def infer(u, t):
+    if len(u) < 1:
+        u = "https://lf3-static.bytednsdoc.com/obj/eden-cn/pbovhozuha/screenshot-20240923-164458.png"
+    if len(t) < 1:
+        t = "请将图里文字转成markdown"
     messages = [
         {
             "role": "user",
             "content": [
                 {
                     "type": "image",
-                    "image": "https://lf3-static.bytednsdoc.com/obj/eden-cn/pbovhozuha/screenshot-20240923-164458.png",
+                    "image": u,
+                },
+                {
+                    "type": "text",
+                    "text": t,
                 },
-                {"type": "text", "text": n},
             ],
         }
     ]
@@ -50,5 +55,12 @@ def infer(n):
     return output_text
     
 
-demo = gr.Interface(fn=infer, inputs=gr.Text(), outputs=gr.Text())
+demo = gr.Interface(
+    fn=infer,
+    inputs=[
+        gr.Text(label="url"),
+        gr.Text(label="text"),
+    ],
+    outputs=gr.Text(),
+)
 demo.launch()