Skip to content

QUICKSTART.md

Latest commit

 

History

History
110 lines (68 loc) · 3 KB

QUICKSTART.md

File metadata and controls

110 lines (68 loc) · 3 KB
Raw

Quickstart

This document gives the shortest practical path through the repository.

Before You Start

Complete setup first:

Quickstart A: Validate ADS Batch Configuration

This is the best first command for a new user because it checks the config structure before creating any designs.

From the repository root:

python .\parallel_version\batch_processor.py validate-config --config .\parallel_version\config_examples\batch_config_pdk.json

Before running it, edit:

  • parallel_version/config_examples/batch_config_pdk.json

At minimum, set:

  • pdk_config.pdk_dir
  • pdk_config.pdk_tech_dir
  • substrate

If the config validates, you have the repository and environment wired together correctly.

Quickstart B: Run the Small Parallel ADS Flow

From the repository root:

python .\parallel_version\batch_processor.py process-config --config .\parallel_version\config_examples\batch_config_pdk.json

When prompted with Proceed with execution? [Y/n]:, answer Y.

This will:

  1. create or open the target workspace
  2. create the target library
  3. create designs from JSON files in config_examples/json_layout/
  4. run EM simulations
  5. export results to the configured output directory

Recommended first-time behavior:

  • keep max_workers small
  • keep batch_size small
  • use the sample json_layout/ directory first

Quickstart C: Build HDF5 From Simulation Results

After the ADS flow produces .sNp files, switch to the ML environment and create an HDF5 dataset.

From the repository root:

python .\Data_process\HDF5_create\create_hdf5.py --json_dir .\parallel_version\config_examples\json_layout --snp_dir .\parallel_version\batch_results --output_dir .\Data_process\HDF5_create\datasets

This groups compatible layouts and S-parameter files into HDF5 datasets. Only valid Touchstone files with readable frequency data are included.

Quickstart D: Train A Model

After you have an HDF5 dataset, stay in the ML environment and train a checkpoint.

From Pytorch_Model/src/:

python train.py --hdf5_path "<path-to-your-dataset.h5>" --epochs 5 --batch_size 16 --checkpoint_name best_model_2ch_2port_16x16.pth

If you prefer to keep datasets under Pytorch_Model/data/, point --hdf5_path there explicitly.

Quickstart E: Verify A Trained Model

Stay in the ML environment.

From Pytorch_Model/src/:

python tools\verify_model.py --hdf5_path "<path-to-your-dataset.h5>" --model_path ..\models\best_model_2ch_2port_16x16.pth --sample_index 0

What Is Not Included By Default

The repository does not ship large runtime assets in Git history:

  • ADS workspaces
  • batch results
  • HDF5 datasets
  • trained model checkpoints

So a new user should expect to either:

  1. generate data locally, or
  2. add externally stored datasets/checkpoints later

Recommended Reading After Quickstart