agentjido / req_llm / a47489b2d5819649e80a12521c32c25cd6afa60b

49%

push

github

Fix cached token extraction from Google API responses (#192)

Google's API returns cachedContentTokenCount in the usageMetadata
field for both implicit (automatic) and explicit (CachedContent API)
prompt caching. The provider was not extracting this value, causing
cached_tokens to always report 0 even when caching was active.

This fix extracts cachedContentTokenCount and converts it to the
OpenAI-compatible prompt_tokens_details.cached_tokens format for both
google and google-vertex providers using Gemini models.

Closes #191

5 of 6 new or added lines in 1 file covered. (83.33%)

3575 of 7149 relevant lines covered (50.01%)

58.79 hits per line