Added Q6_K kv cache option

Author: thad0ctor

.gitignore

@@ -4,3 +4,7 @@ __pycache__
 .gitignore
 llama.log
 llama.log
+log.txt
+tensor_overrides/Kimi-K2-Instruct-UD-IQ1_S-00001-of-00006_Kimi-K2-Instruct-UD-IQ1_S_kv=q4_0_vv=q4_0_th=24_tb=48_ctx=131072_fa_tensor_params.txt
+tensor_overrides/Kimi-K2-Instruct-UD-IQ1_S-00001-of-00006_Kimi-K2-Instruct-UD-IQ1_S_kv=q4_0_vv=q4_0_th=24_tb=48_ctx=131072_fa_tensor_params_balanced.txt
+tensor_overrides/Kimi-K2-Instruct-UD-IQ1_S-00001-of-00006_Kimi-K2-Instruct-UD-IQ1_S_kv=q4_0_vv=q4_0_th=24_tb=48_ctx=131072_fa_tensor_params_original.txt

llamacpp-server-launcher.py

@@ -1063,15 +1063,15 @@ def _setup_advanced_tab(self, parent):
         ttk.Label(inner, text="KV Cache Type (--cache-type-k):")\
             .grid(column=0, row=r, sticky="w", padx=10, pady=3)
         # Combobox state is "readonly" for selection, but not disabled
-        self.cache_type_k_combo = ttk.Combobox(inner, textvariable=self.cache_type_k, width=10, values=("f16","f32","q8_0","q4_0","q4_1","q5_0","q5_1"), state="readonly")
+        self.cache_type_k_combo = ttk.Combobox(inner, textvariable=self.cache_type_k, width=10, values=("f16","f32","q8_0","q4_0","q4_1","q5_0","q5_1","q6_k"), state="readonly")
         self.cache_type_k_combo.grid(column=1, row=r, sticky="w", padx=5, pady=3); r += 1
         ttk.Label(inner, text="Quantization for KV cache (f16 is default, lower Q=more memory saved)", font=("TkSmallCaptionFont"))\
             .grid(column=2, row=r-1, columnspan=2, sticky="w", padx=5, pady=3);
 
         ttk.Label(inner, text="V Cache Type (--cache-type-v):")\
             .grid(column=0, row=r, sticky="w", padx=10, pady=3)
         # Combobox state is "readonly" for selection, but not disabled
-        self.cache_type_v_combo = ttk.Combobox(inner, textvariable=self.cache_type_v, width=10, values=("f16","f32","q8_0","q4_0","q4_1","q5_0","q5_1"), state="readonly")
+        self.cache_type_v_combo = ttk.Combobox(inner, textvariable=self.cache_type_v, width=10, values=("f16","f32","q8_0","q4_0","q4_1","q5_0","q5_1","q6_k"), state="readonly")
         self.cache_type_v_combo.grid(column=1, row=r, sticky="w", padx=5, pady=3); r += 1
         ttk.Label(inner, text="Quantization for V cache (f16 is default, lower Q=more memory saved)", font=("TkSmallCaptionFont"))\
             .grid(column=2, row=r-1, columnspan=2, sticky="w", padx=5, pady=3);

version

@@ -1 +1 @@
-2025-07-12-2
+2025-07-22-0