LLM Compiler

About

Meta's LLM Compiler is an advanced family of large language models specifically designed for the niche of code optimization. Built on the robust foundation of the Code Llama model, it excels in understanding compiler intermediate representations, assembly, and various optimization techniques. Through a rigorous training regimen involving 546 billion tokens of LLVM-IR and assembly languages, the models have undergone fine-tuning to adeptly navigate compiler behaviors. Available in versions with 7-billion and 13-billion parameters, further fine-tuned iterations, LLM Compiler FTD, enhance capabilities in code size optimization and disassembly tasks. These models are licensed for both research and commercial applications, offering versatile solutions in the realm of compiler optimization 26.