April 4, 20262 min readStandalone Inference Working: Shared Memory and a CPU FallbackThe standalone backend finally produced inference after fixing ABI layout, shared-memory buffers, and a simple CPU fallback.SYCLKernelsABILocal LLMPerformanceRead more→