Using VertexAI with prompt caching?

[ShaharZivanOnvego]

Hello,
I have a fully-functional project that uses Anthropic, and utilizes prompt-caching to improve performance and reduce costs.
To further increase speed, I've decided to try using VertexAI.

The way I create my SystemPrompt object is this way:

        content_dict = {
            "text": prompt,
            "type": "text",
        }

        if use_cache:
            content_dict["cache_control"] = {"type": "ephemeral"}

        assistant_prompt = ChatPromptTemplate.from_messages(
            [
                SystemMessage(content=[content_dict]),
                ("placeholder", "{messages}"),
            ]
        )

This, as mentioned, works perfectly fine with ChatAnthropic, and initializing the message with a dict is actually the only way I managed to get the message to actually be cached.

However, when I try to use the exact same code with ChatAnthropicVertex, I get the following error:

ValueError: System message must be a string, instead was: <class 'list'>

I verified that if I initialize the SystemMessage with content=some_string it works, so the dict is definitely the issue.

This is the code snippet that triggers the error:

    merged_messages = _merge_messages(messages)
    for i, message in enumerate(merged_messages):
        if message.type == "system":
            if i != 0:
                raise ValueError("System message must be at beginning of message list.")
            if not isinstance(message.content, str):
                raise ValueError(
                    "System message must be a string, "
                    f"instead was: {type(message.content)}"
                )
            system_message = message.content
            continue

I tried commenting out this part, but then I get:

anthropic.BadRequestError: Error code: 400 - {'type': 'error', 'error': {'type': 'invalid_request_error', 'message': 'system.0.cache_control: Extra inputs are not permitted'}}

So - definitely enforced by Google as well.

If passing this parameter this way is out of the question... how can I use prompt caching on Vertex? Is it even possible?