
Aristo94/EngramHalo.cpp
Speeds up Qwen 3.8 Flash-Next decoding on AMD Strix Halo with MTP, in a llama.cpp fork. It is a fork of ggml-org/llama.cpp.
Benchmarks
- On Ryzen AI MAX+ 395 / Radeon 8060S (Strix Halo), 96 GB with Qwen3.8 Flash-Next (UD-IQ3_XXS), it decodes at 39.3 tok/s, against 24.4 tok/s for MTP off, same patched build (+61%). Source Note: A 400-token code prompt with a q8_0 KV cache at temperature 0, with the engram table held in RAM and short contexts only; with the table on SSD the same run reads 35.3 tok/s. ROCm only.
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Something wrong or out of date?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.