[FEATURE] Add support for Prompt Caching in AWS Bedrock and Anthropic Models via LiteLLM

[ZeroCool2u]

Feature Area

Core functionality

Is your feature request related to a an existing bug? Please link it here.

NA

Describe the solution you'd like

Especially for workflows that use kickoff_for_each() and kickoff_async() with repetitive context, this would dramatically cut costs by leveraging the support that litellm has built-in for prompt caching. See the following link for more info about built-in prompt caching support from litellm: https://docs.litellm.ai/docs/completion/prompt_caching

Describe alternatives you've considered

Spending the extra money on uncached tokens I guess.

Additional context

No response

Willingness to Contribute

I can test the feature once it's implemented

[Vidit-Ostwal]

This is an interesting feature.
From what I understand by looking at this https://docs.litellm.ai/docs/tutorials/prompt_caching#how-it-works

I wasn't able to find the right way to initialise and send the cache_control_injection_points as a parameter to litellm, this is just a try.

llm = LLM(
    model="openrouter/openai/gpt-4o-mini",
    api_key=OPENROUTER_API_KEY,
    cache_control_injection_points = {
        "location" : "message",
        "role" : "system"
    }
)

Do let me know if this works.