llama.cpp

13bc2527 - vlib-video: unit tests U1/U2/U3+U4 (token count, parser, MRoPE bookkeeping)

vlib-video: unit tests U1/U2/U3+U4 (token count, parser, MRoPE bookkeeping) U2 (tool parser) drives the 21-fixture jsonl through parse_tool_call; U1 (pair packing) and U3+U4 (cumulative MRoPE / pop on rewind) are arithmetic-only stubs for now — the full design-doc tests need a stub clip_ctx and live llama_context which we don't have yet. All three build and pass under ctest. See VIDEO_CONV3D_CPP_DESIGN.md §5.1.