dasllama Runs Speech-to-Text 5x Faster Than llama.cpp

Gaijin Entertainment released dasllama, an inference engine written in daslang, which runs speech-to-text 5x faster than llama.cpp on CPU. Benchmarks measured on 2026-10-21/22 show dasllama 0.6.6 outperforming whisper.cpp on M1 hardware. The system uses a macro-based tuner to automatically select optimal CPU permutations for specific hardware like EPYC and Graviton, achieving 2.3x and 2.8x speedups respectively.