ewc_penalty
Compute Elastic Weight Consolidation's Fisher-weighted quadratic penalty, gradient, and additive multi-task quadratic terms.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- ewc_penalty
- description
- Compute Elastic Weight Consolidation's Fisher-weighted quadratic penalty, gradient, and additive multi-task quadratic terms.
02 资源文件 2
- 脚本scripts/ewc_penalty.py2.0 KB
- 脚本tests/test_ewc_penalty.py1.3 KB
03 SKILL.md 正文
EWC Penalty
Use this skill when training a new task after estimating previous-task Fisher importance. It implements the paper equation lambda / 2 * sum(F_i * (theta_i - theta_star_i)^2) and the matching gradient.
Inputs
- Current parameters
theta. - Previous optimum
theta_star. - Diagonal Fisher vector
fisher. - EWC multiplier
lambda_value.
Outputs
- Scalar EWC penalty.
- Gradient contribution to add to the new-task gradient.
- Optional summed penalty over multiple previous anchors.
Workflow
- Validate matching vector dimensions and nonnegative Fisher entries.
- Compute parameter deltas from the old optimum.
- Apply the paper's
lambda / 2penalty scaling. - Add the gradient
lambda * fisher * deltato the task-B gradient. - Sum penalties for multiple previous tasks when needed.
Validation
Run:
python tests/test_ewc_penalty.py
Limitations
This skill computes the consolidation term only. It must be combined with a task-specific loss in a training harness.
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).