ggml-quants : 1.625 bpw ternary packing for BitNet 1.58b

2024-06-19 12:21:08 -04:00 · 2024-06-19 12:21:08 -04:00 · bd807499f7
commit bd807499f7
parent ac146628e4
11 changed files with 594 additions and 4 deletions
--- a/ggml/include/ggml.h
+++ b/ggml/include/ggml.h
@ -383,6 +383,8 @@ extern "C" {
        GGML_TYPE_F64     = 28,
        GGML_TYPE_IQ1_M   = 29,
        GGML_TYPE_BF16    = 30,
+        GGML_TYPE_Q2_2    = 31,
+        GGML_TYPE_Q1_3    = 32,
        GGML_TYPE_COUNT,
    };