Misc. bug: 2 GB at the start

[tim-lebedkov]

Name and Version

$ /ucrt64/bin/llama-cli.exe --version
ggml_opencl: selected platform: 'AMD Accelerated Parallel Processing'

ggml_opencl: device: 'Caicos (OpenCL 1.2 AMD-APP (1800.11))'
Unsupported GPU: Caicos
ggml_opencl: drop unsupported device.

ggml_opencl: device: 'AMD Ryzen 7 1700X Eight-Core Processor (OpenCL 1.2 AMD-APP (1800.11))'
Unsupported GPU: AMD Ryzen 7 1700X Eight-Core Processor
ggml_opencl: drop unsupported device.
version: 52594 (c7e6113b84)
built with gcc.exe (Rev8, Built by MSYS2 project) 15.2.0 for x86_64-w64-mingw32

Operating systems

Windows

Which llama.cpp modules do you know to be affected?

libllama (core library)

Command line

Problem description & steps to reproduce

Just entering the main() in a program linking to "ggml", "ggml-base", "mtmd", "llama" consumes 2 GB of RAM.
VMMap shows it as 16 blocks 128 MB each. Are these allocations necessary? Can they be postponed to when I really use an LLM?

First Bad Commit

No response

Relevant log output

[tim-lebedkov]

Somebody also reserves 4 GB

[tim-lebedkov]

Hello,

Who are these maintainers exactly? Is one already assigned? How long will it take? Can I do something to speed up the process?

[tim-lebedkov]

The same happens if I just link to "openblas". This has nothing to do with libllama.