From 13a39058d3d8bb8aaaf1a981e90d6edf629dc944 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Johannes=20G=C3=A4=C3=9Fler?= <johannesg@5d6.de>
Date: Sun, 10 Mar 2024 19:50:11 +0100
Subject: [PATCH] quantize: fix F16/F32 downcast to q6_K

---
 llama.cpp | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/llama.cpp b/llama.cpp
index 249442166d24e..68694965b986f 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -11675,7 +11675,7 @@ static ggml_type get_k_quant_type(quantize_state_internal & qs, ggml_type new_ty
                  ftype == LLAMA_FTYPE_MOSTLY_IQ1_S   || ftype == LLAMA_FTYPE_MOSTLY_IQ2_S  || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) {
             new_type = GGML_TYPE_Q5_K;
         }
-        else if (new_type != GGML_TYPE_Q8_0) {
+        else if (new_type != GGML_TYPE_Q8_0 && new_type != GGML_TYPE_F16 && new_type != GGML_TYPE_F32) {
             new_type = GGML_TYPE_Q6_K;
         }
     } else if (name == "token_embd.weight") {