cmake : enable curl by default #12761

ngxson · 2025-04-04T21:56:05Z

When curl was first added to the project, we decided to leave it disabled by default because there was no guarantee that the target system has libcurl installed.

But lot of things changed since then, curl now become part of the overall UX of llama.cpp examples. Most of the pre-built version (including own docker images and release binary) have it enabled, so I think it's now a good time to turn it on by default.

ngxson · 2025-04-04T22:21:32Z

I have no idea how to make the build-linux-cross to work, so I'll disable curl for those (unless someone knows how to fix it)

For windows build, I'll have a look to see how we can setup libcurl without copying the same code everywhere.

ngxson · 2025-04-05T12:20:47Z

examples/run/CMakeLists.txt

+# TODO: avoid copying this code block from common/CMakeLists.txt
+set(LLAMA_RUN_EXTRA_LIBS "")
+if (LLAMA_CURL)
+    find_package(CURL REQUIRED)
+    target_compile_definitions(${TARGET} PUBLIC LLAMA_USE_CURL)
+    include_directories(${CURL_INCLUDE_DIRS})
+    find_library(CURL_LIBRARY curl REQUIRED)
+    set(LLAMA_RUN_EXTRA_LIBS ${LLAMA_RUN_EXTRA_LIBS} ${CURL_LIBRARY})
+endif ()


@ericcurtin FYI, seems like llama-run was not linked against curl on windows, which will result in compilation error (cannot find curl/curl.h header). I made this quick fix which should work for now.

By default, only libcommon is linked against curl.

Thanks @ngxson FWIW we are moving away from using llama-run in RamaLama, I wrote a kinda client server thing that automatically terminates both processes on completion instead (the experience is similar to llama-run but better). I was starting to dislike the duplication of efforts in llama-run and llama-server. So I want to try and consolidate on llama-server.

It's a neat tool to test llama-server with all kinds of models, I'm almost ready to push the full version of it to main in RamaLama.

I'll still be happy to look at PRs around llama-run, etc. but it probably won't be actively used in RamaLama anymore at least.

ngxson · 2025-04-05T12:41:40Z

@ggerganov This should be ready for review. Btw, I think it will be more convenient to statically link libcurl on windows as curl is not installed at system level. I can give a try in a follow-up PR.

ggerganov · 2025-04-06T08:45:34Z

CMakeLists.txt

@@ -168,6 +168,10 @@ add_subdirectory(src)
 # utils, programs, examples and tests
 #

+if (NOT LLAMA_BUILD_EXAMPLES OR NOT LLAMA_BUILD_COMMON)


Suggested change

if (NOT LLAMA_BUILD_EXAMPLES OR NOT LLAMA_BUILD_COMMON)

if (NOT LLAMA_BUILD_EXAMPLES AND NOT LLAMA_BUILD_COMMON)

Also add a CMake log message when this branch is executed.

On second thought, I think this condition should be just if NOT LLAMA_BUILD_COMMON, because we cannot build (most of the) examples without libcommon

Fixed in 533e0dc

ggerganov · 2025-04-06T08:51:36Z

Would the Windows binaries in the releases now have curl support after this change?

slaren · 2025-04-06T15:46:09Z

.github/workflows/build-linux-cross.yml

@@ -19,12 +19,15 @@ jobs:
          sudo apt-get install -y --no-install-recommends \
                  build-essential \
                  gcc-14-riscv64-linux-gnu \
-                  g++-14-riscv64-linux-gnu
+                  g++-14-riscv64-linux-gnu \
+                  libcurl4-openssl-dev


Does this work?

Suggested change

libcurl4-openssl-dev

libcurl4-openssl-dev:riscv64

Thanks, it works. I tested on my fork: https://github.com/ngxson/llama.cpp/actions/runs/14295363304/job/40061820312

ngxson · 2025-04-06T18:20:39Z

Would the Windows binaries in the releases now have curl support after this change?

Yeah that's a good question, I thought that we always copy the libcurl-x64.dll to the release artifact, but turns out we don't (we only use libcurl-x64.dll on server CI). For now, I'll add this to the windows release.

slaren · 2025-04-06T20:10:48Z

Redistributing curl may also require including the copyright notice.

ngxson · 2025-04-06T20:16:28Z

Yeah I think we're also missing some licenses for 3rd-party deps like httplib or json.hpp (also license for linenoise doesn't seem to be included in windows build, it's only included on linux builds)

I think what would be nicer is to package these licenses directly into the binary, and allow them to be shown via a CLI switch like --license

For now, to keep this PR simple, I'll simply create the license file on windows build via cmakefile

ngxson · 2025-04-06T21:19:52Z

@slaren I added a script in cmakelists to copy all licenses into build/bin if it's running on github actions: 02257b7

Now I have another problem, ./build-xcframework.sh doesn't seem to respect the cmake config. Unfortunately I don't know who to work with xcode. @ggerganov Do you have any idea how to fix this?

ggerganov · 2025-04-07T04:49:29Z

Now I have another problem, ./build-xcframework.sh doesn't seem to respect the cmake config.

This works on my end:

diff --git a/build-xcframework.sh b/build-xcframework.sh
index 2ce3939c4..1b9091d28 100755
--- a/build-xcframework.sh
+++ b/build-xcframework.sh
@@ -399,6 +399,7 @@ cmake -B build-ios-sim -G Xcode \
     -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=iphonesimulator \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
     -S .
 cmake --build build-ios-sim --config Release -- -quiet
 
@@ -411,6 +412,7 @@ cmake -B build-ios-device -G Xcode \
     -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=iphoneos \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
     -S .
 cmake --build build-ios-device --config Release -- -quiet
 
@@ -421,6 +423,7 @@ cmake -B build-macos -G Xcode \
     -DCMAKE_OSX_ARCHITECTURES="arm64;x86_64" \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
     -S .
 cmake --build build-macos --config Release -- -quiet
 
@@ -434,6 +437,7 @@ cmake -B build-visionos -G Xcode \
     -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=xros \
     -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
     -S .
 cmake --build build-visionos --config Release -- -quiet
 
@@ -447,6 +451,7 @@ cmake -B build-visionos-sim -G Xcode \
     -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=xrsimulator \
     -DCMAKE_C_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="-D_XOPEN_SOURCE=700 ${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
     -S .
 cmake --build build-visionos-sim --config Release -- -quiet
 
@@ -462,6 +467,7 @@ cmake -B build-tvos-sim -G Xcode \
     -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=appletvsimulator \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
     -S .
 cmake --build build-tvos-sim --config Release -- -quiet
 
@@ -476,6 +482,7 @@ cmake -B build-tvos-device -G Xcode \
     -DCMAKE_XCODE_ATTRIBUTE_SUPPORTED_PLATFORMS=appletvos \
     -DCMAKE_C_FLAGS="${COMMON_C_FLAGS}" \
     -DCMAKE_CXX_FLAGS="${COMMON_CXX_FLAGS}" \
+    -DLLAMA_CURL=OFF \
     -S .
 cmake --build build-tvos-device --config Release -- -quiet

ngxson · 2025-04-07T09:04:09Z

Oh ok thanks, no idea why earlier I couldn't find the correct thing in build-xcframework.sh to be fixed, I think I opened the wrong file 🥲

slaren · 2025-04-07T09:10:41Z

CMakeLists.txt

+#
+# copy the license files
+#
+
+# Check if running in GitHub Actions
+if(DEFINED ENV{GITHUB_ACTIONS} AND "$ENV{GITHUB_ACTIONS}" STREQUAL "true")
+    message(STATUS "Running inside GitHub Actions - copying license files")
+
+    # Copy all files from licenses/ to build/bin/
+    file(GLOB LICENSE_FILES "${CMAKE_SOURCE_DIR}/licenses/*")
+    foreach(LICENSE_FILE ${LICENSE_FILES})
+        get_filename_component(FILENAME ${LICENSE_FILE} NAME)
+        configure_file(${LICENSE_FILE} "${CMAKE_BINARY_DIR}/bin/${FILENAME}" COPYONLY)
+    endforeach()
+endif()


This is a hack, but I guess it is ok for now. We could look into using CPack to generate the packages, then these files would be added with install and would be automatically included in all packages.

ngxson · 2025-04-07T12:27:06Z

I have no idea why merging this also affect opening PRs. But if you see compilation error regarding Could NOT find CURL, simply rebase or merge master into your local branch and it will fix the problem.

Reference: ggml-org/llama.cpp#12761

* master: (123 commits) cuda : add f32 to bf16 copy op (ggml-org#12806) llava: improve clip_ctx destructor to not memleak load_image_size (ggml-org#12834) llama : fix FA when KV cache is not used (i.e. embeddings) (ggml-org#12825) server : fix thread.join() on exit (ggml-org#12831) llava: add more helper functions to check projector types in clip context (ggml-org#12824) arg : Including limits file on AIX (ggml-org#12822) server : webui : Improve Chat Input with Auto-Sizing Textarea (ggml-org#12785) Revert "sycl:remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor" (ggml-org#12812) gguf-py : support lazy tensor splitting (ggml-org#12809) llama : Support llama 4 text-only (ggml-org#12791) opencl: better identify Adreno GPU (ggml-org#12760) hellaswag: display estimated score confidence interval (ggml-org#12797) cuda : fix HIP and MUSA BF16 (#0) sync : ggml ggml : simplify Arm fp16 CPU logic (ggml/1177) CUDA: don't convert BF16 weights to FP32 (ggml/1174) cpu: move all the operators into a separate c++ file (except mul_mat) (ggml/1167) sycl: remove redundant memcopy in function ggml_backend_sycl_buffer_set_tensor (ggml-org#12734) ci : no curl on ggml-ci (ggml-org#12796) cmake : enable curl by default (ggml-org#12761) ... # Conflicts: # common/arg.cpp # common/common.cpp # common/common.h

cmake : enable curl by default

6080f6e

ngxson requested a review from ggerganov April 4, 2025 21:56

github-actions bot added build Compilation issues examples devops improvements to build systems and github actions server labels Apr 4, 2025

ngxson added 3 commits April 5, 2025 00:03

no curl if no examples

6455796

fix build

2cc893b

fix build-linux-cross

793074c

ngxson added 10 commits April 5, 2025 12:27

add windows-setup-curl

2238e70

fix

707f2b9

shell

79509cc

fix path

21c423e

fix windows-latest-cmake*

9bf42cd

run: include_directories

a8a7ed9

LLAMA_RUN_EXTRA_LIBS

a963705

sycl: no llama_curl

04eddb8

no test-arg-parser on windows

64faafc

clarification

1c1c2c7

github-actions bot added testing Everything test related SYCL https://en.wikipedia.org/wiki/SYCL - GPU programming language labels Apr 5, 2025

ngxson commented Apr 5, 2025

View reviewed changes

ericcurtin approved these changes Apr 5, 2025

View reviewed changes

ggerganov approved these changes Apr 6, 2025

View reviewed changes

ggerganov requested a review from slaren April 6, 2025 08:51

slaren reviewed Apr 6, 2025

View reviewed changes

ngxson added 3 commits April 6, 2025 20:22

try riscv64 / arm64

aec8b46

windows: include libcurl inside release binary

26166a4

add msg

533e0dc

fix mac / ios / android build

43a7eec

github-actions bot added the android Issues specific to Android label Apr 6, 2025

ngxson added 4 commits April 6, 2025 22:45

will this fix xcode?

c45a9c4

try clearing the cache

fef3248

add bunch of licenses

02257b7

revert clear cache

b5aafd6

ngxson added 3 commits April 7, 2025 10:24

fix xcode

ca0f7c4

fix xcode (2)

40fe8f2

fix typo

d926a54

ngxson requested a review from slaren April 7, 2025 08:57

slaren approved these changes Apr 7, 2025

View reviewed changes

ngxson mentioned this pull request Apr 7, 2025

changelog : libllama API #9289

Open

ngxson merged commit bd3f59f into ggml-org:master Apr 7, 2025
51 checks passed

felladrin added a commit to felladrin/MiniSearch that referenced this pull request Apr 7, 2025

Add curl package and disable LLAMA_CURL in cmake build options

3c85f51

Reference: ggml-org/llama.cpp#12761

thetick1234567890 mentioned this pull request Apr 9, 2025

Add libcurl support for vulkan x64 zip package #12368

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

cmake : enable curl by default #12761

cmake : enable curl by default #12761

ngxson commented Apr 4, 2025 •

edited

Loading

ngxson commented Apr 4, 2025

ngxson Apr 5, 2025

ericcurtin Apr 5, 2025

ngxson commented Apr 5, 2025

ggerganov Apr 6, 2025

ngxson Apr 6, 2025

ngxson Apr 6, 2025

ggerganov commented Apr 6, 2025

slaren Apr 6, 2025

ngxson Apr 6, 2025

ngxson commented Apr 6, 2025 •

edited

Loading

slaren commented Apr 6, 2025

ngxson commented Apr 6, 2025 •

edited

Loading

ngxson commented Apr 6, 2025

ggerganov commented Apr 7, 2025

ngxson commented Apr 7, 2025

slaren Apr 7, 2025

ngxson commented Apr 7, 2025

	if (NOT LLAMA_BUILD_EXAMPLES OR NOT LLAMA_BUILD_COMMON)
	if (NOT LLAMA_BUILD_EXAMPLES AND NOT LLAMA_BUILD_COMMON)

cmake : enable curl by default #12761

cmake : enable curl by default #12761

Conversation

ngxson commented Apr 4, 2025 • edited Loading

ngxson commented Apr 4, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ngxson commented Apr 5, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ggerganov commented Apr 6, 2025

Choose a reason for hiding this comment

Choose a reason for hiding this comment

ngxson commented Apr 6, 2025 • edited Loading

slaren commented Apr 6, 2025

ngxson commented Apr 6, 2025 • edited Loading

ngxson commented Apr 6, 2025

ggerganov commented Apr 7, 2025

ngxson commented Apr 7, 2025

Choose a reason for hiding this comment

ngxson commented Apr 7, 2025

ngxson commented Apr 4, 2025 •

edited

Loading

ngxson commented Apr 6, 2025 •

edited

Loading

ngxson commented Apr 6, 2025 •

edited

Loading