firecoperana 0b126b2ca6 Fix prompt tokenization issue during prompt processing (#1008) ...

* Find common tokens between prompt and cache
Fix wrong context size usage for mtmd
Use start position of common part
server: handle context shift

* Add size check for inexact match

* Change

---------

Co-authored-by: firecoperana <firecoperana>

2025-11-26 10:34:26 +01:00

64 KiB

Raw Blame History

View Raw