From 4c5f077bd3e808637ad343ffc536a1959b743065 Mon Sep 17 00:00:00 2001
From: Krrish Dholakia <krrishdholakia@gmail.com>
Date: Wed, 19 Jun 2024 18:47:08 -0700
Subject: [PATCH] fix(vertex_ai.py): check if message length > 0 before merging

---
 litellm/llms/vertex_ai.py                       | 7 +++++--
 litellm/tests/test_amazing_vertex_completion.py | 4 ++--
 2 files changed, 7 insertions(+), 4 deletions(-)

diff --git a/litellm/llms/vertex_ai.py b/litellm/llms/vertex_ai.py
index 392be24a857..1dbd93048d3 100644
--- a/litellm/llms/vertex_ai.py
+++ b/litellm/llms/vertex_ai.py
@@ -337,7 +337,7 @@ def _gemini_convert_messages_with_history(messages: list) -> List[ContentType]:
                 _parts: List[PartType] = []
                 for element in messages[msg_i]["content"]:
                     if isinstance(element, dict):
-                        if element["type"] == "text":
+                        if element["type"] == "text" and len(element["text"]) > 0:
                             _part = PartType(text=element["text"])
                             _parts.append(_part)
                         elif element["type"] == "image_url":
@@ -345,7 +345,10 @@ def _gemini_convert_messages_with_history(messages: list) -> List[ContentType]:
                             _part = _process_gemini_image(image_url=image_url)
                             _parts.append(_part)  # type: ignore
                 user_content.extend(_parts)
-            else:
+            elif (
+                isinstance(messages[msg_i]["content"], str)
+                and len(messages[msg_i]["content"]) > 0
+            ):
                 _part = PartType(text=messages[msg_i]["content"])
                 user_content.append(_part)
 
diff --git a/litellm/tests/test_amazing_vertex_completion.py b/litellm/tests/test_amazing_vertex_completion.py
index 4edc580a9b6..742e0388019 100644
--- a/litellm/tests/test_amazing_vertex_completion.py
+++ b/litellm/tests/test_amazing_vertex_completion.py
@@ -570,7 +570,6 @@ async def test_gemini_pro_vision(provider, sync_mode):
         # Google counts the prompt tokens for us, we should ensure we use the tokens from the orignal response
         assert prompt_tokens == 263  # the gemini api returns 263 to us
 
-        assert False
     except litellm.RateLimitError as e:
         pass
     except Exception as e:
@@ -1164,6 +1163,7 @@ def test_gemini_pro_vision_async():
             resp = await litellm.acompletion(
                 model="vertex_ai/gemini-pro-vision",
                 messages=[
+                    {"role": "system", "content": ""},
                     {
                         "role": "user",
                         "content": [
@@ -1175,7 +1175,7 @@ def test_gemini_pro_vision_async():
                                 },
                             },
                         ],
-                    }
+                    },
                 ],
             )
             print("async response gemini pro vision")