AWS Bedrock Cached Tokens Info in Langfuse

[fabriciojoc]

Hello there! First of all, congrats for this awesome project!

I've been using it connected to AWS Bedrock (Claude 3.7) and Langfuse 3.29.0. I recently enabled prompt caching, but I do not see any cache info in Langfuse (like cached_creation_input_tokens or cache_read_input_tokens as documented by AWS https://aws.amazon.com/blogs/machine-learning/effectively-use-prompt-caching-on-amazon-bedrock/).

Am I missing a configuration or is it not supported by LiteLLM yet? I noticed in the integration code that you are getting only the prompt_tokens and completion_tokens, shouldn't it include the cached tokens there?

Thanks!

[krrishdholakia]

Does langfuse have fields we can send this to?

We do track this already and use it for calculating costs.

[fabriciojoc]

Looks like the usage_details object can include the fields for that (such as cached_creation_input_tokens and cache_read_input_tokens), according to this page.