ikawrakow/ik_llama.cpp
1
0
Fork 0
mirror of https://github.com/ikawrakow/ik_llama.cpp.git synced 2026-05-01 11:51:53 +00:00
Code Issues Packages Projects Releases Wiki Activity

Update printour of allowed quantized KV-cache combinations

Browse Source
This commit is contained in:
Iwan Kawrakow
2024-10-21 08:32:52 +03:00
parent 67acecce76
commit ca7e403946
1 changed files with 5 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
ggml/src/ggml-cuda/fattn-common.cuh
View File

@@ -591,9 +591,10 @@ static void on_no_fattn_vec_case(const int D) {
fprintf(stderr, "Unsupported KV type combination for head_size 128.\n");
fprintf(stderr, "Supported combinations:\n");
fprintf(stderr, " - K == q4_0, V == q4_0, 4.50 BPV\n");
fprintf(stderr, " - K == q8_0, V == iq4_nl, 6.50 BPV\n");
fprintf(stderr, " - K == q8_0, V == q8_0, 8.50 BPV\n");
fprintf(stderr, " - K == f16, V == f16, 16.00 BPV\n");
fprintf(stderr, "Compile with GGML_CUDA_FA_ALL_QUANTS for all combinations of q4_0, q4_1, q5_0, q5_1, q8_0, and f16.\n");
fprintf(stderr, "Compile with GGML_CUDA_FA_ALL_QUANTS for all combinations of q4_0, q4_1, iq4_nl, q5_0, q5_1, q8_0, and f16.\n");
GGML_ABORT("fatal error");
} else {
fprintf(stderr, "Unsupported KV type combination for head_size 256.\n");