ik_llama_opt/ggml
Alex ab6d816818
CUDA DSA: fix v_offset for quantized K/V caches (#2387)
v_offset is used as a column index into the dequantized f16 buffer, but was
computed as a half-pointer difference. That is only correct for f16 K/V. With
q8_0 (34 bytes per 32 elements) it yields the wrong column, so the sparse
attention path reads V from the wrong positions and generation degenerates.
2026-08-31 18:35:17 +02:00
..
cmake Merge mainline llama.cpp (#3) 2024-07-27 07:55:01 +02:00
include Quantization fudge factors (#2361) 2026-08-27 17:35:04 +02:00
src CUDA DSA: fix v_offset for quantized K/V caches (#2387) 2026-08-31 18:35:17 +02:00
.gitignore Merge mainline llama.cpp (#3) 2024-07-27 07:55:01 +02:00
CMakeLists.txt Chunked experts (CPU) (#2202) 2026-07-30 13:16:02 +03:00