Skip to content

Navigation Menu

Sign in
Appearance settings

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit 117cbb2

Browse filesBrowse files
abetlenabetlen
committed
feat: Update llama.cpp
1 parent 04959f1 commit 117cbb2
Copy full SHA for 117cbb2

File tree

Expand file treeCollapse file tree

3 files changed

+44
-34
lines changed
Filter options
Expand file treeCollapse file tree

3 files changed

+44
-34
lines changed

‎CMakeLists.txt

Copy file name to clipboardExpand all lines: CMakeLists.txt
+37-32Lines changed: 37 additions & 32 deletions
Original file line numberDiff line numberDiff line change
@@ -5,9 +5,28 @@ project(llama_cpp)
55
option(LLAMA_BUILD "Build llama.cpp shared library and install alongside python package" ON)
66
option(LLAVA_BUILD "Build llava shared library and install alongside python package" ON)
77

8+
if(SKBUILD_STATE STREQUAL "editable")
9+
# Install into the source directory
10+
# Temporary fix for https://github.com/scikit-build/scikit-build-core/issues/374
11+
set(LLAMA_CPP_PYTHON_INSTALL_DIR ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp)
12+
else()
13+
set(LLAMA_CPP_PYTHON_INSTALL_DIR ${SKBUILD_PLATLIB_DIR}/llama_cpp)
14+
endif()
15+
816
if (LLAMA_BUILD)
917
set(BUILD_SHARED_LIBS "On")
1018

19+
set(CMAKE_SKIP_BUILD_RPATH FALSE)
20+
21+
# When building, don't use the install RPATH already
22+
# (but later on when installing)
23+
set(CMAKE_BUILD_WITH_INSTALL_RPATH FALSE)
24+
25+
# Add the automatically determined parts of the RPATH
26+
# which point to directories outside the build tree to the install RPATH
27+
set(CMAKE_INSTALL_RPATH_USE_LINK_PATH TRUE)
28+
set(CMAKE_INSTALL_RPATH "${LLAMA_CPP_PYTHON_INSTALL_DIR}")
29+
1130
# Building llama
1231
if (APPLE AND NOT CMAKE_SYSTEM_PROCESSOR MATCHES "arm64")
1332
# Need to disable these llama.cpp flags on Apple x86_64,
@@ -24,31 +43,26 @@ if (LLAMA_BUILD)
2443

2544
add_subdirectory(vendor/llama.cpp)
2645
install(
27-
TARGETS llama
28-
LIBRARY DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
29-
RUNTIME DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
30-
ARCHIVE DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
31-
FRAMEWORK DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
32-
RESOURCE DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
46+
TARGETS llama
47+
LIBRARY DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
48+
RUNTIME DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
49+
ARCHIVE DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
50+
FRAMEWORK DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
51+
RESOURCE DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
3352
)
34-
# Temporary fix for https://github.com/scikit-build/scikit-build-core/issues/374
3553
install(
36-
TARGETS llama
37-
LIBRARY DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
38-
RUNTIME DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
39-
ARCHIVE DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
40-
FRAMEWORK DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
41-
RESOURCE DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
54+
TARGETS ggml
55+
LIBRARY DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
56+
RUNTIME DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
57+
ARCHIVE DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
58+
FRAMEWORK DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
59+
RESOURCE DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
4260
)
4361
# Workaround for Windows + CUDA https://github.com/abetlen/llama-cpp-python/issues/563
4462
if (WIN32 AND (LLAMA_CUDA OR LLAMA_CUBLAS))
4563
install(
4664
FILES $<TARGET_RUNTIME_DLLS:llama>
47-
DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
48-
)
49-
install(
50-
FILES $<TARGET_RUNTIME_DLLS:llama>
51-
DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
65+
DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
5266
)
5367
endif()
5468

@@ -71,20 +85,11 @@ if (LLAMA_BUILD)
7185
endif()
7286
install(
7387
TARGETS llava_shared
74-
LIBRARY DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
75-
RUNTIME DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
76-
ARCHIVE DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
77-
FRAMEWORK DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
78-
RESOURCE DESTINATION ${SKBUILD_PLATLIB_DIR}/llama_cpp
79-
)
80-
# Temporary fix for https://github.com/scikit-build/scikit-build-core/issues/374
81-
install(
82-
TARGETS llava_shared
83-
LIBRARY DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
84-
RUNTIME DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
85-
ARCHIVE DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
86-
FRAMEWORK DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
87-
RESOURCE DESTINATION ${CMAKE_CURRENT_SOURCE_DIR}/llama_cpp
88+
LIBRARY DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
89+
RUNTIME DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
90+
ARCHIVE DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
91+
FRAMEWORK DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
92+
RESOURCE DESTINATION ${LLAMA_CPP_PYTHON_INSTALL_DIR}
8893
)
8994
endif()
9095
endif()

‎llama_cpp/llama_cpp.py

Copy file name to clipboardExpand all lines: llama_cpp/llama_cpp.py
+6-1Lines changed: 6 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -273,6 +273,7 @@ def byref(obj: CtypesCData, offset: Optional[int] = None) -> CtypesRef[CtypesCDa
273273
# LLAMA_VOCAB_TYPE_SPM = 1, // LLaMA tokenizer based on byte-level BPE with byte fallback
274274
# LLAMA_VOCAB_TYPE_BPE = 2, // GPT-2 tokenizer based on byte-level BPE
275275
# LLAMA_VOCAB_TYPE_WPM = 3, // BERT tokenizer based on WordPiece
276+
# LLAMA_VOCAB_TYPE_UGM = 4, // T5 tokenizer based on Unigram
276277
# };
277278
LLAMA_VOCAB_TYPE_NONE = 0
278279
"""For models without vocab"""
@@ -282,6 +283,8 @@ def byref(obj: CtypesCData, offset: Optional[int] = None) -> CtypesRef[CtypesCDa
282283
"""GPT-2 tokenizer based on byte-level BPE"""
283284
LLAMA_VOCAB_TYPE_WPM = 3
284285
"""BERT tokenizer based on WordPiece"""
286+
LLAMA_VOCAB_TYPE_UGM = 4
287+
"""T5 tokenizer based on Unigram"""
285288

286289

287290
# // pre-tokenization types
@@ -302,6 +305,7 @@ def byref(obj: CtypesCData, offset: Optional[int] = None) -> CtypesRef[CtypesCDa
302305
# LLAMA_VOCAB_PRE_TYPE_DBRX = 13,
303306
# LLAMA_VOCAB_PRE_TYPE_SMAUG = 14,
304307
# LLAMA_VOCAB_PRE_TYPE_PORO = 15,
308+
# LLAMA_VOCAB_PRE_TYPE_VIKING = 16,
305309
# };
306310
LLAMA_VOCAB_PRE_TYPE_DEFAULT = 0
307311
LLAMA_VOCAB_PRE_TYPE_LLAMA3 = 1
@@ -319,6 +323,7 @@ def byref(obj: CtypesCData, offset: Optional[int] = None) -> CtypesRef[CtypesCDa
319323
LLAMA_VOCAB_PRE_TYPE_DBRX = 13
320324
LLAMA_VOCAB_PRE_TYPE_SMAUG = 14
321325
LLAMA_VOCAB_PRE_TYPE_PORO = 15
326+
LLAMA_VOCAB_PRE_TYPE_VIKING = 16
322327

323328

324329
# // note: these values should be synchronized with ggml_rope
@@ -2317,7 +2322,7 @@ def llama_n_threads_batch(ctx: llama_context_p, /) -> int:
23172322
...
23182323

23192324

2320-
# // Set whether the model is in embeddings model or not
2325+
# // Set whether the model is in embeddings mode or not
23212326
# // If true, embeddings will be returned but logits will not
23222327
# LLAMA_API void llama_set_embeddings(struct llama_context * ctx, bool embeddings);
23232328
@ctypes_function("llama_set_embeddings", [llama_context_p_ctypes, ctypes.c_bool], None)

‎vendor/llama.cpp

Copy file name to clipboard

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.