FlexLLM: A Composable HLS Library for Rapid LLM Accelerator Design

https://github.com/Crazy-James26/FlexLLM

FlexLLM is a composable High-Level Synthesis (HLS) library for rapidly building hybrid temporal–spatial accelerators for Large Language Models (LLMs). It provides parameterized module templates, optimized memory-access/dataflow components, and a complete quantization suite, enabling FPGA-based LLM systems to be built with minimal manual engineering effort.