Add eps as a parameter to ne_norm function

[aahouzi]

Type of Change

• eps parameter is hardcoded to 1e-5 in the ne_norm function, and many models in the convert script look for rms_norm_eps while the model doesn't use RMS normalization or have bad naming compared to config.json file.
• This PR fixes the above issues for all models, and suggests a single norm_eps parameter for both RMS normalization or classic normalization.

Description

• Same as above

Expected Behavior & Potential Risk

• N/A

How has this PR been tested?

• Works fine for all models

Dependency Change?

• N/A

[a32543254]

LGTM

[a32543254]

thanks for you contribution.
After this PR, we could adopt Rms eps and Layernorm eps as one args.
could you kindly also check the Format Scan ?

[aahouzi]

@a32543254 all checks are passing now. However, it's little bit weird, since when I apply clang-format on the project directory for my modified files, it didn't pass the clangformat test. When I introduced the commit 7ea5623 , it passed. Actually, I expect the .clang-format file to put everything in place or am I missing smthg here ?

[a32543254]

@a32543254 all checks are passing now. However, it's little bit weird, since when I apply clang-format on the project directory for my modified files, it didn't pass the clangformat test. When I introduced the commit 7ea5623 , it passed. Actually, I expect the .clang-format file to put everything in place or am I missing smthg here ?

you can simple use clang-format.py in repo to auto check the clang format.
Everything is done by now. We will merge this pr soon