[Model] SiglipVisionModel ported from transformers (#6942)

Co-authored-by: Roger Wang <ywang@roblox.com>

examples/offline_inference_vision_language.py

@@ -65,7 +65,8 @@ def run_phi3v(question):
 # PaliGemma
 def run_paligemma(question):
 
-    prompt = question
+    # PaliGemma has special prompt format for VQA
+    prompt = "caption en"
     llm = LLM(model="google/paligemma-3b-mix-224")
 
     return llm, prompt