Closed MichaelZhangBH closed 2 months ago
I have the same issue running a 64 bit binary on Windows 10 compiled with VS2022. AVX and AVX2 are enabled and using CPU backend. The bark and encodec models were converted with the python scripts to f16. No errors output when converting the models or running main.exe, it just crashes at this function call.
@jzeiber @MichaelZhangBH Hello both! Thanks for opening this issue.
@MichaelZhangBH Could you post the entire traceback?
Hello @PABannier , great to have you look into this. I will put the call stack here.
ggml.dll!__chkstk() Line 109 Unknown
> ggml.dll!ggml_allocr_new_from_buffer(ggml_backend_buffer * buffer) Line 304 C
main.exe!bark_model_load(std::basic_ifstream<char,std::char_traits<char>> & fin, gpt_model & model, int n_gpu_layers, bark_verbosity_level verbosity) Line 882 C++
main.exe!bark_load_model_from_file(const std::string & fname, bark_model * model, int n_gpu_layers, bark_verbosity_level verbosity) Line 1029 C++
main.exe!bark_load_model(const std::string & model_path, bark_verbosity_level verbosity) Line 1066 C++
main.exe!main(int argc, char * * argv) Line 34 C++
[External Code]
Also, BTW, is there any plan to upgrade the to some of the more recent commits of ggml.
@MichaelZhangBH Yes, it is on my todo list. I'll try to do that in the next few days!
Hello @PABannier can you please upload proper working binary files for Windows to Release?
I was able to get it working. The default stack size in Visual Studio is 1MB. I changed it to 100MB, and now bark successfully generates output.
@jzeiber Thank you for trying! Could you open a PR and add your fix in a CMakeLists.txt
file? Using if(MSVC)
and increasing the stack size should do the trick.
Closing as this is fixed with #163
Built with master branch on Windows 11. Same error w/wo CUDA.
Unhandled exception at 0x00007FFB0C804477 (ggml.dll) in main.exe: 0xC00000FD: Stack overflow (parameters: 0x0000000000000001, 0x0000009026403000).