Update llama.cpp

Author: abetlen

Makefile

@@ -30,6 +30,12 @@ build.metal:
 build.vulkan:
 	CMAKE_ARGS="-DLLAMA_VULKAN=on" python3 -m pip install --verbose -e .
 
+build.kompute:
+	CMAKE_ARGS="-DLLAMA_KOMPUTE=on" python3 -m pip install --verbose -e .
+
+build.sycl:
+	CMAKE_ARGS="-DLLAMA_SYCL=on" python3 -m pip install --verbose -e .
+
 build.sdist:
 	python3 -m build --sdist
 

llama_cpp/llama_cpp.py

@@ -187,6 +187,7 @@ def _load_shared_library(lib_base_name: str):
 #     LLAMA_FTYPE_MOSTLY_IQ2_XS        = 20, // except 1d tensors
 #     LLAMA_FTYPE_MOSTLY_Q2_K_S        = 21, // except 1d tensors
 #     LLAMA_FTYPE_MOSTLY_Q3_K_XS       = 22, // except 1d tensors
+#     LLAMA_FTYPE_MOSTLY_IQ3_XXS       = 23, // except 1d tensors
 
 #     LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file
 # };
@@ -211,6 +212,7 @@ def _load_shared_library(lib_base_name: str):
 LLAMA_FTYPE_MOSTLY_IQ2_XS = 20
 LLAMA_FTYPE_MOSTLY_Q2_K_S = 21
 LLAMA_FTYPE_MOSTLY_Q3_K_XS = 22
+LLAMA_FTYPE_MOSTLY_IQ3_XXS = 23
 LLAMA_FTYPE_GUESSED = 1024
 
 # enum llama_rope_scaling_type {