masked_program_embedding
Build BAR-style masked target embeddings and universal tanh-bounded programs while preserving protected target-domain data.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- masked_program_embedding
- description
- Build BAR-style masked target embeddings and universal tanh-bounded programs while preserving protected target-domain data.
02 资源文件 3
03 SKILL.md 正文
Masked Program Embedding
Use this skill when implementing adversarial reprogramming or a BAR proxy experiment that must embed smaller target-domain arrays in a larger source-domain input and add a universal program only outside the protected region.
Do not use it for ordinary data augmentation or for perturbations that are allowed to alter the target sample content.
Inputs
- A batch of same-shaped numeric target samples.
- A source canvas shape with the same rank as each sample.
- An embedding offset, or
Noneto center the sample. - Program parameters
Wwith the source canvas shape.
Outputs
- Embedded batch
X. - Binary mask
M, with0on target data and1on programmable cells. - Program
P = tanh(W*M). - Programmed batch
X + P.
Workflow
- Validate that every target sample fits inside the source canvas.
- Place every target sample at the same offset in a zero canvas.
- Construct a mask that protects exactly the embedded target region.
- Apply the universal tanh-bounded program outside that region.
- Verify protected values are unchanged before passing data to a black-box model.
Validation
Run python tests/test_masked_program_embedding.py or validate the skill tree with validate_skill_tree.py --run-tests.
Limitations
This skill does not choose label mappings or optimize W; it only provides the input transformation contract used by BAR.
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).