Invert image and text

app.py

@@ -61,10 +61,10 @@ def pixtral_inference(
     chat = [
         {
             "role": "user",
-            "content": [
+            "content": [{"type": "image", "url": image_to_base64(i[0])} for i in images]
+            + [
                 {"type": "text", "content": text},
-            ]
-            + [{"type": "image", "url": image_to_base64(i[0])} for i in images],
+            ],
         }
     ]
 
@@ -76,7 +76,7 @@ def pixtral_inference(
         return_tensors="pt",
     ).to(model.device)
     print("GENERATING")
-    generate_ids = model.generate(**inputs, max_new_tokens=500)
+    generate_ids = model.generate(**inputs, max_new_tokens=256)
     print("GENERATING DONE")
     print("BATCH DECODE")
     output = processor.batch_decode(
@@ -84,7 +84,7 @@ def pixtral_inference(
     )[0]
     print("BATCH DECODE DONE")
     print(output)
-    result = output.text
+    result = output
     return result