Token & cost optimization
2 min · Lobsters
dasllama Uses Compile-Time AST Macros for 5x Faster CPU Speech and LLM Inference
dasllama adapts Gaijin Entertainment's daslang game scripting language into an inference engine that runs automatic speech recognition 5x faster than whisper.cpp on CPU. By replacing hand-tuned assembly kernels with compile-time AST macros and autotuning sidecars, it maximizes hardware utilization without writing external C++ or custom shaders.
Why it matters. You can run low-latency local speech recognition and LLM inference alongside resource-heavy rendering using lightweight compile-time CPU autotuning.