Kalsa
Engines / GenerelSchwerz / GenerelSchwerz/llama.cpp

GenerelSchwerz/llama.cpp

NVIDIAForkMIT

Changes how llama.cpp places its cache on NVIDIA GPUs, and decodes Qwen3.6 35B faster than stock. It is a fork of ggml-org/llama.cpp.

Benchmarks

  • On RTX 5070 Ti 16 GB with Qwen3.6 35B (Q4_K_M), it decodes at 146.77 tok/s, against 64.08 tok/s for llama.cpp (2.3×). Source Note: A one-run observation, not a repeat median. The author placed stock so its measured peak GPU memory is higher than the fork's (11,751 MiB against 11,443 MiB), and says cache placement changes the numerical path, so the output check shows no obvious degeneration but does not establish equal answer quality.

Listing

Listed: Passed all seven checks.

Security review

Not reviewed yet.

Install

To install it, follow the project's own instructions on GitHub. The engine runs with your user's permissions. macOS doesn't sandbox it, so install only engines you trust.

Something wrong or out of date? Write to help@kalsa.io, or ask us to remove this page.