pgd_adversarial_training_loop
Run a deterministic PGD adversarial training loop that updates model parameters on adversarial examples and logs loss traces.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- pgd_adversarial_training_loop
- description
- Run a deterministic PGD adversarial training loop that updates model parameters on adversarial examples and logs loss traces.
02 资源文件 2
- 脚本scripts/adversarial_train.py4.0 KB
- 脚本tests/test_adversarial_train.py937 B
03 SKILL.md 正文
PGD Adversarial Training Loop
When To Use
Use this skill for the outer minimization step in a Madry-style robust optimization recovery. It is designed for bounded experiments where the model can generate adversarial examples with the current parameters and then update parameters on those examples.
Do not use it to claim robust training if no PGD examples are generated or no trainable parameters change.
Inputs
- Training examples and labels.
- A trainable model with parameter gradients.
- PGD attack configuration: epsilon, step size, steps, restarts, and seed.
- Optimizer configuration: learning rate and epochs.
Outputs
- Updated model parameters.
- Training trace with
loss_before,loss_after,params_before, andparams_after. - Mechanism checks for PGD generation and optimizer execution.
Workflow
- Generate PGD adversarial examples using the current model.
- Compute adversarial loss and parameter gradients.
- Apply a gradient-descent update to model weights and bias.
- Re-evaluate adversarial loss after the update.
- Save trace fields that prove parameters changed and loss was measured before and after.
Validation
Run:
python scripts/adversarial_train.py --self-test
python tests/test_adversarial_train.py
Limitations
The included script is a deterministic reduced proxy using a tiny logistic classifier. Full MNIST/CIFAR recovery requires a real deep-learning runtime and dataset.
04 同仓库更多 Skills
passnet-triton-opt
Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.
passnet-solve
END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.
passnet-skill
PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.
passnet-pattern-fusion
Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.
passnet-orchestrate
Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.
passnet-feedback
Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).