Compile bug: ggml_vulkan:compiling shaders aborted on qualcomm adreno 750 gpu

[AndreaChiChengdu]

Git commit

b4483

Operating systems

Linux

GGML backends

Vulkan

Problem description & steps to reproduce

I used Termux on a Qualcomm SM8650 phone to configure the Vulkan build environment according to the Ubuntu guide, and the build was successful.

However, when I tried to load the LLaMA 3 8B model using llama-cli , I encountered the issue described in the title: ggml_vulkan compiling shaders aborted .

First Bad Commit

No response

Compile command

./bin/llama-cli -m llama3-8b.gguf -t 4 -n 100 -p "xxxxx"

Relevant log output

please check the failed log photo above
thanks

[jeffbolznv]

Please try running with the Vulkan Validation Layers enabled.

[AndreaChiChengdu]

Please try running with the Vulkan Validation Layers enabled.

hello sir ，Is there any documentation for that? I tried -ngl 33 and it had the same problem

[jeffbolznv]

Not sure about this specific environment, but it's probably roughly:

• Install the Vulkan SDK (see https://vulkan.lunarg.com/doc/sdk/1.4.304.0/linux/getting_started_ubuntu.html)
• export VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation

[samkoesnadi]

Not sure about this specific environment, but it's probably roughly:

• Install the Vulkan SDK (see https://vulkan.lunarg.com/doc/sdk/1.4.304.0/linux/getting_started_ubuntu.html)
• export VK_INSTANCE_LAYERS=VK_LAYER_KHRONOS_validation

I just tried this solution. It unfortunately does not work. Still aborted like before. I have opened an issue that provides more details: #11327

[jeffbolznv]

Please try again at TOT (after #11406). Shaders are compiled on-demand and may dodge whatever issue you're running into here.

[github-actions]

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