Can not release memory

[wuxianjun666]

in ggml_extend.hpp, the destructor function ~GGMLModule(),
~GGMLModule() noexcept(false){
LOG_DEBUG("BEGIN TO SUPER RELEASE .........");
free_params_buffer();
LOG_DEBUG("SUPER RELEASE FINISHED.........");
}
It always gets stuck here. because of this function: free_params_buffer();
there ara many things succeed GGMLModule(), such as LoraModel, AutoEncoderKL, UNetModel, FrozenCLIPEmbedderWithCustomWords

[wuxianjun666]

can not delete sd_ctx->sd;
when delete sd_ctx->sd, it should delete GGMLModule many times, because there are many struct succeed GGMLModule,

but always gets stuck here,

[leejet]

Can you provide more information, such as the operating system you're using, the compiler version, the command-line arguments you're using, and the corresponding output?

[wuxianjun666]

Can you provide more information, such as the operating system you're using, the compiler version, the command-line arguments you're using, and the corresponding output?

I run in windows10, and the command is "cmake .. and cmake --build ."
the output is:

Check for working C compiler: C:/Program Files (x86)/Microsoft Visual Studio/2019/Community/VC/Tools/MSVC/14.29.30133/bin/Hostx64/x64/cl.exe

Check for working CUDA compiler: C:/Program Files/NVIDIA GPU Computing Toolkit/CUDA/v11.6/bin/nvcc.exe

the error happened in when i finished generate image and save it, it begin to delete ctx->sd, but the program gets stuck.
The stuck position is in the function ~GGMLModule(), because the unet, vae, clip all succeed the struct GGMLModule,
so the function ~GGMLModule() will run many time. so i add some print in the function "void free_params_buffer()"
when run the code "LOG_DEBUG("params_ctx != NULL, BEGIN TO free .........");" the program gets stuck. there is the screenshot

[wuxianjun666]

Can you provide more information, such as the operating system you're using, the compiler version, the command-line arguments you're using, and the corresponding output?

So I had to eliminate this function so it would run fine and not get stuck.

[snowyu]

Also this error occurs on Ubuntu 22.04.3 whenever Lora is used.

build/bin/sd -m models/Stable-diffusion/v1-5-pruned-emaonly.safetensors  --lora-model-dir models/Lora -v  -p "a cute cat<lora:lcm-lora-sdv1-5:1>" --cfg-scale 1 --steps 4
[DEBUG] ggml_extend.hpp:601  - unet compute buffer size: 560.71 MB
  |==================================================| 4/4 - 2.54s/it
[INFO ] stable-diffusion.cpp:1247 - sampling completed, taking 11.80s
[INFO ] stable-diffusion.cpp:1255 - generating 1 latent images completed, taking 11.94s
[INFO ] stable-diffusion.cpp:1257 - decoding 1 latents
[DEBUG] ggml_extend.hpp:601  - vae compute buffer size: 1665.00 MB
[DEBUG] stable-diffusion.cpp:1076 - computing vae [mode: DECODE] graph completed, taking 8.78s
[INFO ] stable-diffusion.cpp:1267 - latent 1 decoded, taking 8.78s
[INFO ] stable-diffusion.cpp:1271 - decode_first_stage completed, taking 8.78s
[INFO ] stable-diffusion.cpp:1288 - txt2img completed in 20.78s
save result image to 'output.png'
Segment fault(core dumped)

[wuxianjun666]

yes，when i do not use lora, it is normal

[wuxianjun666]

I found out what the problem is, in lora.hpp, comment out（注释） this line of code and it works!

[leejet]

This should be fixed.