Anonymous
6/19/2025, 5:29:08 PM
No.105641526
>>105641209
There are no partial cache rewrites. It either has to be a perfect match or it writes an entirely new cache.
In your example, if I understand it correctly, nothing would be counted for the cheaper cache price. It would just create a new cache at 1.25x/2x (5m/1h TTL) the cost for the full 1280 tokens.
There is a technical option to insert cache breakpoints, as the official documentation states. https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching#putting-it-all-together-multiple-cache-breakpoints
That would allow you to do something like you described, but that is not supported in the standard caching function of SillyTavern.
There are no partial cache rewrites. It either has to be a perfect match or it writes an entirely new cache.
In your example, if I understand it correctly, nothing would be counted for the cheaper cache price. It would just create a new cache at 1.25x/2x (5m/1h TTL) the cost for the full 1280 tokens.
There is a technical option to insert cache breakpoints, as the official documentation states. https://docs.anthropic.com/en/docs/build-with-claude/prompt-caching#putting-it-all-together-multiple-cache-breakpoints
That would allow you to do something like you described, but that is not supported in the standard caching function of SillyTavern.