Evan Jones

Save and restore prompt evaluation state for much faster startup times

5

Hi! I decided to take a stab at leveraging the new get / set state APIs to cache initial prompt evaluation in `main`. On my M2 at least, this feature...

llama : add grammar-based sampling

19

Inspired by #1397 and [grantslatton's CFG work](https://github.com/grantslatton/llama.cpp/commit/007e26a99d485007f724957fa8545331ab8d50c3), this adds an API that takes a serialized context-free grammar to guide and constrain sampling. Also adds a sample Backus-Naur form (BNF)-like syntax...

examples : generate JSON according to schema

5

**Depends on #1773** (to test this, merge that first) Adds a Python script that converts a JSON schema into the grammar format from #1773. This allows generating JSON according to...