From 8503d59ee44bc30b0d030cceb5e17590b334730d Mon Sep 17 00:00:00 2001
From: Ed Addario <eaddario@hotmail.com>
Date: Sat, 13 Sep 2025 11:49:18 +0100
Subject: [PATCH] Increase IQ options

---
 src/llama-quant.cpp | 11 ++++++++++-
 1 file changed, 10 insertions(+), 1 deletion(-)

diff --git a/src/llama-quant.cpp b/src/llama-quant.cpp
index 955e6c12fe..41fd819f86 100644
--- a/src/llama-quant.cpp
+++ b/src/llama-quant.cpp
@@ -641,12 +641,21 @@ static std::unordered_map<std::string, ggml_type> target_bpw_type(
 
     constexpr ggml_type iq_quants[] = {
         GGML_TYPE_IQ1_S,
+        GGML_TYPE_IQ2_XXS,
+        GGML_TYPE_IQ2_XS,
         GGML_TYPE_IQ2_S,
         GGML_TYPE_IQ3_S,
         GGML_TYPE_IQ4_XS,
+        GGML_TYPE_IQ4_NL,
         GGML_TYPE_Q5_K,
         GGML_TYPE_Q6_K,
-        GGML_TYPE_Q8_0
+        GGML_TYPE_Q8_0,
+        // TODO: find better way to handle F16/BF16
+#ifdef GGML_USE_METAL
+        GGML_TYPE_F16
+#else
+        GGML_TYPE_BF16
+#endif
     };
 
     constexpr double epsilon = 1e-12;