From cae691c3911781882fbc884abc80da828d9a97cb Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?=E6=88=90=E9=BE=99?= Date: Fri, 9 May 2025 15:52:50 +0800 Subject: [PATCH] [LLM4C] Create the Initial Optimization Framework for LLM4Compiler. (1) Self-refine: https://github.com/madaan/self-refine.git (2) Initialize the LLM4Compiler(AI-native) Framework for Code Optimization Agent; (3) Add the initial README to introduce the New Member of AI4Compiler. --- LLM4Compiler/README.md | 5 +++++ LLM4Compiler/self-refine | 1 + 2 files changed, 6 insertions(+) create mode 100644 LLM4Compiler/README.md create mode 160000 LLM4Compiler/self-refine diff --git a/LLM4Compiler/README.md b/LLM4Compiler/README.md new file mode 100644 index 00000000..10c83f4c --- /dev/null +++ b/LLM4Compiler/README.md @@ -0,0 +1,5 @@ +# LLM4Compiler - VecTrans: Iterative Refinement LLM Optimization Agent Framework for Auto-vectorization. + +[Vision]: For the next-generation innovation of Compiler and Code Optimization in the Era of LLM. + +* The developed VecTrans Agent Framework is initially based on self-refine(https:selfrefine.info/). However, we extended more feedback sources to be integrated into LLM, and enhance the code optimization capability for LLVM-based Compilers. diff --git a/LLM4Compiler/self-refine b/LLM4Compiler/self-refine new file mode 160000 index 00000000..9a206d41 --- /dev/null +++ b/LLM4Compiler/self-refine @@ -0,0 +1 @@ +Subproject commit 9a206d41e5d2d0c241bb441f41eeadb945afaa55 -- Gitee