Help, please! Source code compile.

目标平台：910B4 CANN 8.5.1
软件版本：
FlagTree： 0.6.0+triton3.2（commit 951af6322）
AscendNPU-IR：commit 5a3921f87
flir：triton_v3.3.x（commit 2756c17）

想要使用源码编译 llvm 在使用编译后产物编译FlagTree，使用的llvm commit ：a66376b0
可以正常编译出FlagTree，但是使用FlagGems推理时报错：
[flagos_infer_46840.log](https://github.com/user-attachments/files/31742072/flagos_infer_46840.log)

使用说明文档中预编译的llvm-a66376b0-ubuntu-aarch64-python311-compat 可以正常启动 完成推理。请问是llvm版本的问题嘛？你们有针对llvm进行修改嘛？

文档中提供的 wget https://baai-cp-web.ks3-cn-beijing.ksyuncs.com/trans/llvm-a66376b0-ubuntu-aarch64-python311-compat_v0.3.0.tar.gz 对应的源代码能提供嘛？

万分感谢！

Start by reviewing flagos_infer_46840.log and comparing the source-built LLVM at commit a66376b0 with the documented prebuilt llvm-a66376b0-ubuntu-aarch64-python311-compat_v0.3.0 package. Check the referenced FlagTree, AscendNPU-IR, and flir commits for build assumptions. Done means determining whether the builds differ and documenting the required source or build instructions.

Written by the indexing model from the issue text.