Update docs

Author: KuangYu

docs/user_guide/4.5CustomTorch.md

@@ -4,6 +4,8 @@ This module is designed to support pytorch machine learning potential (MLP) kern
 It aims to perform top-down finetuning for MLP models writen in torch
 A full simple example can be found in `examples/torch_kernel/torch_mlp_frontend`
 To use this module, you need to install both `torch` and [torch2jax](https://github.com/samuela/torch2jax)
+(TIP: Do not use `pip install torch2jax` directly, `git clone` the repo and do `pip install .` there, which
+gives you the newer version that is compatible with new versions of JAX)
 
 NOTE: currently we DO NOT support differentiating losses related to force and virial!