modify the README for ML

Author: tyh0123

Exec/README_md.pytorch

@@ -1,36 +1,89 @@
-# PyTorch (libtorch) Download and Setup
+# README_pytorch.md
+## Running a MagneX / AMReX Application with a PyTorch Model (LibTorch + TorchScript)
 
-This guide downloads the libtorch CUDA 11.8 C++ distribution and unzips it in the same directory that contains `MagneX`, then renames the extracted folder to `libtorch_cuda`.
+This guide explains how to run an AMReX-based application (e.g., MagneX) that evaluates a pre-trained PyTorch model using the LibTorch C++ API.
 
-## Steps
+---
 
-1. Change to the parent directory of `MagneX`:
+## Overview
+
+The typical workflow is:
+
+1. Initialize an AMReX `MultiFab` with input data (e.g., different `dt` values).
+2. Copy the `MultiFab` data into a `torch::Tensor`.
+3. Load a TorchScript model (`model.pt`).
+4. Run inference on CPU or GPU.
+5. Copy the model outputs back into a `MultiFab`.
+6. Write AMReX plotfiles for visualization.
+
+The model **must** be exported as TorchScript (`model.pt`).  
+Do not use a Python-only `.pth` checkpoint.
+
+---
+
+## Requirements
+
+- MagneX (or another AMReX-based application)
+- A TorchScript model file:
+  ```
+  model.pt
+  ```
+- LibTorch (CUDA build recommended for GPU runs)
+- CUDA version tested: **13.0**
+- LibTorch version tested: **2.9.1 + cu126**
+
+---
+
+## Directory Layout (Recommended)
+
+Place LibTorch in the same parent directory as MagneX:
 
-```bash
-cd <MAGNEX_PARENT_DIR>
 ```
+/path/to/work/
+    MagneX/
+    libtorch_cuda/
+```
+
+---
+
+## Step 1: Download LibTorch (CUDA cu126)
 
-2. Download the libtorch archive with `wget`:
+From your working directory:
 
 ```bash
-wget https://download.pytorch.org/libtorch/cu118/libtorch-cxx11-abi-shared-with-deps-2.7.1%2Bcu118.zip
+wget https://download.pytorch.org/libtorch/cu126/libtorch-shared-with-deps-2.9.1%2Bcu126.zip
+unzip libtorch-shared-with-deps-2.9.1+cu126.zip
+mv libtorch libtorch_cuda
 ```
 
-3. Unzip the archive in the same directory:
+This creates:
+
+```
+libtorch_cuda/
+```
+
+If needed, adjust your build system so it can find:
 
-```bash
-unzip libtorch-cxx11-abi-shared-with-deps-2.7.1+cu118.zip
 ```
+libtorch_cuda/include
+libtorch_cuda/lib
+```
+
+---
+
+## Step 2: Build MagneX with CUDA + ML Enabled
 
-4. Rename the extracted folder to `libtorch_cuda`:
+Go to your MagneX directory:
 
 ```bash
-mv libtorch libtorch_cuda
+cd /path/to/work/MagneX/Exec
+make -j4 USE_CUDA=TRUE USE_ML=TRUE
 ```
 
-## Result
+This should produce an executable similar to:
 
-After the steps above, you should have:
+```
+main2d.gnu.MPI.CUDA.ex
+```
 
-- `<MAGNEX_PARENT_DIR>/MagneX`
-- `<MAGNEX_PARENT_DIR>/pytorch_cuda`
+The exact name depends on compiler and MPI configuration.