merge: TurboQuant KV cache types onto upstream master (MTP built-in)

2026-05-18 01:52:24 +03:00
parent 726704a160
commit b34ac655f3
72 changed files with 18946 additions and 257 deletions
@@ -407,6 +407,9 @@ const std::vector<ggml_type> kv_cache_types = {
    GGML_TYPE_IQ4_NL,
    GGML_TYPE_Q5_0,
    GGML_TYPE_Q5_1,
+    GGML_TYPE_TURBO2_0,
+    GGML_TYPE_TURBO3_0,
+    GGML_TYPE_TURBO4_0,
 };

 static ggml_type kv_cache_type_from_str(const std::string & s) {