1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
|
diff -rU1 ollama.orig/llm/generate/gen_linux.sh ollama.nocuda.norocm/llm/generate/gen_linux.sh
--- ollama.orig/llm/generate/gen_linux.sh 2024-05-12 11:46:20.026663875 +0200
+++ ollama.nocuda.norocm/llm/generate/gen_linux.sh 2024-05-12 11:58:20.653330156 +0200
@@ -53,3 +53,3 @@
fi
-COMMON_CMAKE_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=off -DLLAMA_AVX=on -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off"
+COMMON_CMAKE_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=on"
source $(dirname $0)/gen_common.sh
@@ -66,3 +66,3 @@
CMAKE_TARGETS="--target llama --target ggml"
- CMAKE_DEFS="-DBUILD_SHARED_LIBS=off -DLLAMA_NATIVE=off -DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
+ CMAKE_DEFS="-DBUILD_SHARED_LIBS=on -DLLAMA_NATIVE=on ${CMAKE_DEFS}"
BUILD_DIR="../build/linux/${ARCH}_static"
@@ -95,3 +95,3 @@
- COMMON_CPU_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=off"
+ COMMON_CPU_DEFS="-DCMAKE_POSITION_INDEPENDENT_CODE=on -DLLAMA_NATIVE=on"
if [ -z "${OLLAMA_CPU_TARGET}" -o "${OLLAMA_CPU_TARGET}" = "cpu" ]; then
@@ -101,3 +101,3 @@
init_vars
- CMAKE_DEFS="${COMMON_CPU_DEFS} -DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_FMA=off -DLLAMA_F16C=off ${CMAKE_DEFS}"
+ CMAKE_DEFS="${COMMON_CPU_DEFS} ${CMAKE_DEFS}"
BUILD_DIR="../build/linux/${ARCH}/cpu"
@@ -143,71 +143,76 @@
-# If needed, look for the default CUDA toolkit location
-if [ -z "${CUDA_LIB_DIR}" ] && [ -d /usr/local/cuda/lib64 ]; then
- CUDA_LIB_DIR=/usr/local/cuda/lib64
-fi
-
-# If needed, look for CUDA on Arch Linux
-if [ -z "${CUDA_LIB_DIR}" ] && [ -d /opt/cuda/targets/x86_64-linux/lib ]; then
- CUDA_LIB_DIR=/opt/cuda/targets/x86_64-linux/lib
-fi
-
-# Allow override in case libcudart is in the wrong place
-if [ -z "${CUDART_LIB_DIR}" ]; then
- CUDART_LIB_DIR="${CUDA_LIB_DIR}"
-fi
-
-if [ -d "${CUDA_LIB_DIR}" ]; then
- echo "CUDA libraries detected - building dynamic CUDA library"
- init_vars
- CUDA_MAJOR=$(ls "${CUDA_LIB_DIR}"/libcudart.so.* | head -1 | cut -f3 -d. || true)
- if [ -n "${CUDA_MAJOR}" ]; then
- CUDA_VARIANT=_v${CUDA_MAJOR}
- fi
- if [ "${ARCH}" == "arm64" ]; then
- echo "ARM CPU detected - disabling unsupported AVX instructions"
-
- # ARM-based CPUs such as M1 and Tegra do not support AVX extensions.
- #
- # CUDA compute < 6.0 lacks proper FP16 support on ARM.
- # Disabling has minimal performance effect while maintaining compatibility.
- ARM64_DEFS="-DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_CUDA_F16=off"
- fi
- # Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
- if [ -n "${OLLAMA_CUSTOM_CUDA_DEFS}" ]; then
- echo "OLLAMA_CUSTOM_CUDA_DEFS=\"${OLLAMA_CUSTOM_CUDA_DEFS}\""
- CMAKE_CUDA_DEFS="-DLLAMA_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${OLLAMA_CUSTOM_CUDA_DEFS}"
- echo "Building custom CUDA GPU"
- else
- CMAKE_CUDA_DEFS="-DLLAMA_CUDA=on -DLLAMA_CUDA_FORCE_MMQ=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
- fi
- CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS}"
- BUILD_DIR="../build/linux/${ARCH}/cuda${CUDA_VARIANT}"
- EXTRA_LIBS="-L${CUDA_LIB_DIR} -lcudart -lcublas -lcublasLt -lcuda"
- build
-
- # Carry the CUDA libs as payloads to help reduce dependency burden on users
- #
- # TODO - in the future we may shift to packaging these separately and conditionally
- # downloading them in the install script.
- DEPS="$(ldd ${BUILD_DIR}/bin/ollama_llama_server )"
- for lib in libcudart.so libcublas.so libcublasLt.so ; do
- DEP=$(echo "${DEPS}" | grep ${lib} | cut -f1 -d' ' | xargs || true)
- if [ -n "${DEP}" -a -e "${CUDA_LIB_DIR}/${DEP}" ]; then
- cp "${CUDA_LIB_DIR}/${DEP}" "${BUILD_DIR}/bin/"
- elif [ -e "${CUDA_LIB_DIR}/${lib}.${CUDA_MAJOR}" ]; then
- cp "${CUDA_LIB_DIR}/${lib}.${CUDA_MAJOR}" "${BUILD_DIR}/bin/"
- elif [ -e "${CUDART_LIB_DIR}/${lib}" ]; then
- cp -d ${CUDART_LIB_DIR}/${lib}* "${BUILD_DIR}/bin/"
- else
- cp -d "${CUDA_LIB_DIR}/${lib}*" "${BUILD_DIR}/bin/"
- fi
- done
- compress
-fi
+### CUDA stuff deactivated by patch:
-if [ -z "${ROCM_PATH}" ]; then
- # Try the default location in case it exists
- ROCM_PATH=/opt/rocm
-fi
+# # If needed, look for the default CUDA toolkit location
+# if [ -z "${CUDA_LIB_DIR}" ] && [ -d /usr/local/cuda/lib64 ]; then
+# CUDA_LIB_DIR=/usr/local/cuda/lib64
+# fi
+#
+# # If needed, look for CUDA on Arch Linux
+# if [ -z "${CUDA_LIB_DIR}" ] && [ -d /opt/cuda/targets/x86_64-linux/lib ]; then
+# CUDA_LIB_DIR=/opt/cuda/targets/x86_64-linux/lib
+# fi
+#
+# # Allow override in case libcudart is in the wrong place
+# if [ -z "${CUDART_LIB_DIR}" ]; then
+# CUDART_LIB_DIR="${CUDA_LIB_DIR}"
+# fi
+#
+# if [ -d "${CUDA_LIB_DIR}" ]; then
+# echo "CUDA libraries detected - building dynamic CUDA library"
+# init_vars
+# CUDA_MAJOR=$(ls "${CUDA_LIB_DIR}"/libcudart.so.* | head -1 | cut -f3 -d. || true)
+# if [ -n "${CUDA_MAJOR}" ]; then
+# CUDA_VARIANT=_v${CUDA_MAJOR}
+# fi
+# if [ "${ARCH}" == "arm64" ]; then
+# echo "ARM CPU detected - disabling unsupported AVX instructions"
+#
+# # ARM-based CPUs such as M1 and Tegra do not support AVX extensions.
+# #
+# # CUDA compute < 6.0 lacks proper FP16 support on ARM.
+# # Disabling has minimal performance effect while maintaining compatibility.
+# ARM64_DEFS="-DLLAMA_AVX=off -DLLAMA_AVX2=off -DLLAMA_AVX512=off -DLLAMA_CUDA_F16=off"
+# fi
+# # Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
+# if [ -n "${OLLAMA_CUSTOM_CUDA_DEFS}" ]; then
+# echo "OLLAMA_CUSTOM_CUDA_DEFS=\"${OLLAMA_CUSTOM_CUDA_DEFS}\""
+# CMAKE_CUDA_DEFS="-DLLAMA_CUDA=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES} ${OLLAMA_CUSTOM_CUDA_DEFS}"
+# echo "Building custom CUDA GPU"
+# else
+# CMAKE_CUDA_DEFS="-DLLAMA_CUDA=on -DLLAMA_CUDA_FORCE_MMQ=on -DCMAKE_CUDA_ARCHITECTURES=${CMAKE_CUDA_ARCHITECTURES}"
+# fi
+# CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} ${ARM64_DEFS} ${CMAKE_CUDA_DEFS}"
+# BUILD_DIR="../build/linux/${ARCH}/cuda${CUDA_VARIANT}"
+# EXTRA_LIBS="-L${CUDA_LIB_DIR} -lcudart -lcublas -lcublasLt -lcuda"
+# build
+#
+# # Carry the CUDA libs as payloads to help reduce dependency burden on users
+# #
+# # TODO - in the future we may shift to packaging these separately and conditionally
+# # downloading them in the install script.
+# DEPS="$(ldd ${BUILD_DIR}/bin/ollama_llama_server )"
+# for lib in libcudart.so libcublas.so libcublasLt.so ; do
+# DEP=$(echo "${DEPS}" | grep ${lib} | cut -f1 -d' ' | xargs || true)
+# if [ -n "${DEP}" -a -e "${CUDA_LIB_DIR}/${DEP}" ]; then
+# cp "${CUDA_LIB_DIR}/${DEP}" "${BUILD_DIR}/bin/"
+# elif [ -e "${CUDA_LIB_DIR}/${lib}.${CUDA_MAJOR}" ]; then
+# cp "${CUDA_LIB_DIR}/${lib}.${CUDA_MAJOR}" "${BUILD_DIR}/bin/"
+# elif [ -e "${CUDART_LIB_DIR}/${lib}" ]; then
+# cp -d ${CUDART_LIB_DIR}/${lib}* "${BUILD_DIR}/bin/"
+# else
+# cp -d "${CUDA_LIB_DIR}/${lib}*" "${BUILD_DIR}/bin/"
+# fi
+# done
+# compress
+#
+# fi
+
+### ROCm stuff deactivated by patch:
+
+# if [ -z "${ROCM_PATH}" ]; then
+# # Try the default location in case it exists
+# ROCM_PATH=/opt/rocm
+# fi
@@ -220,33 +225,35 @@
-if [ -d "${ROCM_PATH}" ]; then
- echo "ROCm libraries detected - building dynamic ROCm library"
- if [ -f ${ROCM_PATH}/lib/librocblas.so.*.*.????? ]; then
- ROCM_VARIANT=_v$(ls ${ROCM_PATH}/lib/librocblas.so.*.*.????? | cut -f5 -d. || true)
- fi
- init_vars
- CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DLLAMA_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
- # Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
- if [ -n "${OLLAMA_CUSTOM_ROCM_DEFS}" ]; then
- echo "OLLAMA_CUSTOM_ROCM_DEFS=\"${OLLAMA_CUSTOM_ROCM_DEFS}\""
- CMAKE_DEFS="${CMAKE_DEFS} ${OLLAMA_CUSTOM_ROCM_DEFS}"
- echo "Building custom ROCM GPU"
- fi
- BUILD_DIR="../build/linux/${ARCH}/rocm${ROCM_VARIANT}"
- EXTRA_LIBS="-L${ROCM_PATH}/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ -Wl,-rpath,\$ORIGIN/../../rocm/ -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu"
- build
+### ROCm stuff deactivated by patch:
- # Record the ROCM dependencies
- rm -f "${BUILD_DIR}/bin/deps.txt"
- touch "${BUILD_DIR}/bin/deps.txt"
- for dep in $(ldd "${BUILD_DIR}/bin/ollama_llama_server" | grep "=>" | cut -f2 -d= | cut -f2 -d' ' | grep -e rocm -e amdgpu -e libtinfo ); do
- echo "${dep}" >> "${BUILD_DIR}/bin/deps.txt"
- done
- # bomb out if for some reason we didn't get a few deps
- if [ $(cat "${BUILD_DIR}/bin/deps.txt" | wc -l ) -lt 8 ] ; then
- cat "${BUILD_DIR}/bin/deps.txt"
- echo "ERROR: deps file short"
- exit 1
- fi
- compress
-fi
+# if [ -d "${ROCM_PATH}" ]; then
+# echo "ROCm libraries detected - building dynamic ROCm library"
+# if [ -f ${ROCM_PATH}/lib/librocblas.so.*.*.????? ]; then
+# ROCM_VARIANT=_v$(ls ${ROCM_PATH}/lib/librocblas.so.*.*.????? | cut -f5 -d. || true)
+# fi
+# init_vars
+# CMAKE_DEFS="${COMMON_CMAKE_DEFS} ${CMAKE_DEFS} -DLLAMA_HIPBLAS=on -DCMAKE_C_COMPILER=$ROCM_PATH/llvm/bin/clang -DCMAKE_CXX_COMPILER=$ROCM_PATH/llvm/bin/clang++ -DAMDGPU_TARGETS=$(amdGPUs) -DGPU_TARGETS=$(amdGPUs)"
+# # Users building from source can tune the exact flags we pass to cmake for configuring llama.cpp
+# if [ -n "${OLLAMA_CUSTOM_ROCM_DEFS}" ]; then
+# echo "OLLAMA_CUSTOM_ROCM_DEFS=\"${OLLAMA_CUSTOM_ROCM_DEFS}\""
+# CMAKE_DEFS="${CMAKE_DEFS} ${OLLAMA_CUSTOM_ROCM_DEFS}"
+# echo "Building custom ROCM GPU"
+# fi
+# BUILD_DIR="../build/linux/${ARCH}/rocm${ROCM_VARIANT}"
+# EXTRA_LIBS="-L${ROCM_PATH}/lib -L/opt/amdgpu/lib/x86_64-linux-gnu/ -Wl,-rpath,\$ORIGIN/../../rocm/ -lhipblas -lrocblas -lamdhip64 -lrocsolver -lamd_comgr -lhsa-runtime64 -lrocsparse -ldrm -ldrm_amdgpu"
+# build
+#
+# # Record the ROCM dependencies
+# rm -f "${BUILD_DIR}/bin/deps.txt"
+# touch "${BUILD_DIR}/bin/deps.txt"
+# for dep in $(ldd "${BUILD_DIR}/bin/ollama_llama_server" | grep "=>" | cut -f2 -d= | cut -f2 -d' ' | grep -e rocm -e amdgpu -e libtinfo ); do
+# echo "${dep}" >> "${BUILD_DIR}/bin/deps.txt"
+# done
+# # bomb out if for some reason we didn't get a few deps
+# if [ $(cat "${BUILD_DIR}/bin/deps.txt" | wc -l ) -lt 8 ] ; then
+# cat "${BUILD_DIR}/bin/deps.txt"
+# echo "ERROR: deps file short"
+# exit 1
+# fi
+# compress
+# fi
|