[Engine] <llama.cpp-tq3>

[ahmedg4m3s-spec]

Engine / fork name

llama.cpp-tq3

Repo URL

https://github.com/turbo-tan/llama.cpp-tq3

What is it based on?

llama.cpp fork (llama-server compatible)

What's it good for?

this fork should support TQ3_1S KV CACHE which is basically 3.5bits a sweet spot between turbo 3 and turbo4

Prebuilt binaries?

Yes — published on GitHub Releases

Prebuilt asset example (optional)

No response

Supported operating systems

• Windows
• macOS
• Linux

Supported GPUs / accelerators

• NVIDIA (CUDA)
• AMD (ROCm)
• Intel (SYCL)
• Apple Silicon (Metal)
• CPU only

Supported CPU architecture

• x86-64 (x64)
• ARM64 (aarch64 / Apple Silicon)

Anything else?

No response