replace Qwen2-VL with Florence-2 API for caption generation

app.py

@@ -1,3 +1,11 @@
+import requests
+from io import BytesIO
+FLORENCE_URL = "https://api-inference.huggingface.co/models/microsoft/Florence-2-large"
+HF_HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"} if "HF_TOKEN" in locals() or "HF_TOKEN" in globals() else {}
+import requests
+from io import BytesIO
+FLORENCE_URL = "https://api-inference.huggingface.co/models/microsoft/Florence-2-large"
+HF_HEADERS = {"Authorization": f"Bearer {HF_TOKEN}"} if "HF_TOKEN" in locals() or "HF_TOKEN" in globals() else {}
 
 import os
 import torch
@@ -86,7 +94,7 @@ def generate_captions_api(image):
                 ],
                 "max_tokens": 80
             }
-            response = requests.post(
+            response = requests.post(FLORENCE_URL, 
                 
                 headers=HF_HEADERS,
                 json=payload,
@@ -134,7 +142,7 @@ def compute_jina_scores(image, captions):
                 "documents": ["data:image/jpeg;base64," + img_b64],
                 "top_n": 1
             }
-            response = requests.post(
+            response = requests.post(FLORENCE_URL, 
                 JINA_URL,
                 headers=JINA_HEADERS,
                 json=payload,
@@ -232,7 +240,7 @@ def fuse_captions_api(cap1, cap2, dino_labels):
             "temperature"       : 0.1,
             "repetition_penalty": 1.1
         }
-        response = requests.post(
+        response = requests.post(FLORENCE_URL, 
             QWEN_LM_URL,
             headers=HF_HEADERS,
             json=payload,