Bug: Can't build LLAMA_CURL=ON to embed curl on windows x64 build. #9937

Activity

1. AnthonyEmiom

   added

   bug-unconfirmed

   high severityUsed to report high severity bugs in llama.cpp (Malfunctioning hinder important workflow)Used to report high severity bugs in llama.cpp (Malfunctioning hinder important workflow)

   on Oct 18, 2024
2. github-actions

   added

   stale

   on Nov 18, 2024
3. github-actions commented on Dec 2, 2024

   

   github-actionsBot

   on Dec 2, 2024 – with GitHub Actions

   Contributor

   More actions

   This issue was closed because it has been inactive for 14 days since being marked as stale.
4. github-actions

   closed this as completedon Dec 2, 2024
5. gumaerc commented on Feb 3, 2025

   

   gumaerc

   on Feb 3, 2025

   More actions

   I can confirm this is an issue for me as well. I tried locally building CURL in Windows with the newest and older versions. Setting CURL_LIBRARY and CURL_INCLUDE_DIR in my environment did not work. I read that you can use the -DCURL_LIBRARY and -DCURL_INCLUDE_DIR arguments to specify them at the command line and tried that, but the latest Windows release of CURL that you download from them does not include the proper .lib file.

   So, I went and downloaded the source and compiled it. I switched to the winbuild directory and ran nmake /f Makefile.vc mode=static and it compiled the library. I pointed the llama.cpp build at the output, and I got a bunch of errors that look similar to this:

   lld-link: error: undefined symbol: __declspec(dllimport) curl_easy_strerror
   >>> referenced by common.lib(common.cpp.obj):(bool __cdecl curl_perform_with_retry(class std::basic_string<char, struct std::char_traits<char>, class std::allocator<char>> const &, void *, int, int))
   >>> referenced by common.lib(common.cpp.obj):(bool __cdecl curl_perform_with_retry(class std::basic_string<char, struct std::char_traits<char>, class std::allocator<char>> const &, void *, int, int))
   >>> referenced by common.lib(common.cpp.obj):(bool __cdecl curl_perform_with_retry(class std::basic_string<char, struct std::char_traits<char>, class std::allocator<char>> const &, void *, int, int))
   clang++: error: linker command failed with exit code 1 (use -v to see invocation)
   [238/251] Generating index.html.gz.hpp
   ninja: build stopped: subcommand failed.
   
6. randaller commented on May 1, 2025

   

   randaller

   on May 1, 2025

   Last edited by randaller

   More actions

   To build llama.cpp with curl using MSVC:

   Enter llama.cpp git directory and:

   D:
   cd \llama.cpp
   git clone https://github.com/microsoft/vcpkg.git
   cd vcpkg
   bootstrap-vcpkg.bat
   vcpkg.exe install curl:x64-windows
   

   and then build as usual, just adding paths:

   D:
   cd \llama.cpp\build
   
   cmake .. -DGGML_CUDA=ON -DGGML_CUDA_FA_ALL_QUANTS=ON ^
   -DLLAMA_CURL=ON ^
   -DCURL_LIBRARY=D:\llama.cpp\vcpkg\packages\curl_x64-windows\lib\libcurl.lib ^
   -DCURL_INCLUDE_DIR=D:\llama.cpp\vcpkg\packages\curl_x64-windows\include
   
   cmake --build . --config Release
   

   Do not forget to copy \llama.cpp\vcpkg\installed\x64-windows\bin\ libcurl.dll and zlib1.dll to a \llama.cpp folder.

   👍React with 👍6Reacted by Kiu Huang, RoyWu, Raymond Lo, Gontz, jungletsubasa and izogfif

   😄React with 😄2Reacted by Raymond Lo and jungletsubasa

   🎉React with 🎉3Reacted by Raymond Lo, Gontz and jungletsubasa

   😕React with 😕1Reacted by Raymond Lo

   ❤️React with ❤️2Reacted by Anwar Hossain and Raymond Lo

   🚀React with 🚀1Reacted by Raymond Lo

   👀React with 👀1Reacted by Raymond Lo
7. subbu124 commented on May 16, 2025

   

   subbu124

   on May 16, 2025

   Last edited by subbu124

   More actions

   randeller : i tried your way on my laptop with no gpu, all compiled, but cli.exe or server.exe, are not working, nothing happens after command, it shifts to console without doing anything. anythoughts? actually i am not trying to build with curl, but it is giving curl error, so i tried your method, it compiled but facing above issue
8. randaller commented on May 17, 2025

   

   randaller

   on May 17, 2025

   Last edited by randaller

   More actions

   i tried your way on my laptop with no gpu, all compiled, but cli.exe or server.exe, are not working, nothing happens after command, it shifts to console without doing anything. anythoughts

   Did you removed -DGGML_CUDA=ON -DGGML_CUDA_FA_ALL_QUANTS=ON parameters? I am compiling for cuda and just forgot to remove from example. Also check if libcurl and zlib dlls lays in llama-cli.exe folder. For my cuda build parameters there also should be three cudart/cublas libraries, may be this causes error.