In helping to unify the AMD software ecosystem around LLMs on Ryzen AI NPUs and Radeon GPUs, FastFlowLM has been pulled into the ROCm organization. FastFlowLM is now developed under "ROCm" on GitHub. This MIT licensed software continues to support all AMD XDNA2 NPUs and later.
In addition to moving under the ROCm umbrella, FastFlowLM 1.0 adds support for the SmolVLA vision-language-action (VLA) model, flm bench benchmarking improvements, and various bug fixes.
It was only earlier this year that AMD Ryzen AI NPUs became useful on Linux for running LLMs with the FastFlowLM 0.9.35 release having added Linux support. Since then with Lemonade integration and the like, Ryzen AI NPUs are becoming practical on Linux.
The FastFlowLM source code as well as pre-built Ubuntu and Debian binaries, a generic Linux x86_64 binary, and Windows binary can be downloaded from GitHub for those wanting to use it for running LLMs on AMD Ryzen AI NPUs.