minimum_sd_optimizer
Minimise empirical Stein discrepancy objectives with bounded searches and auditable loss traces for recovery experiments.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- minimum_sd_optimizer
- description
- Minimise empirical Stein discrepancy objectives with bounded searches and auditable loss traces for recovery experiments.
02 资源文件 2
03 SKILL.md 正文
Minimum Stein Discrepancy Optimizer
Use this skill when an empirical Stein discrepancy has been defined and a recovery needs to estimate a parameter by minimising that discrepancy. It is especially useful when full stochastic-gradient or Riemannian-gradient infrastructure is unnecessary or unavailable, but the recovery still needs executable evidence of an argmin search.
Inputs
- A callable loss function of a scalar parameter.
- Initial parameter, search bounds, grid size, and optional local refinement radius.
- Metadata about the sample and acceptance threshold.
Outputs
- Best parameter, best loss, initial loss, and loss improvement.
- Search trace with candidate evaluations.
params_beforeandparams_afterfields suitable for reduced recovery validation.
Workflow
- Evaluate the initial parameter and record it as the baseline.
- Evaluate a bounded grid over the supplied interval.
- Optionally refine locally around the best grid point.
- Select the finite candidate with minimum loss.
- Record whether the parameter changed and whether the objective improved.
Validation
Run python tests/test_minimum_sd_optimizer.py or validate the skill tree with tests. Tests include a deterministic quadratic and a monotone trace check.
Limitations
This skill is a bounded deterministic optimiser for recovery evidence. It does not replace the paper's full stochastic Riemannian optimiser for large-scale experiments.
Refinement cycle 1 note
Stress recovery confirmed loss improvement on the deterministic Student-t proxy; keep params_before and params_after in every optimiser trace.
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).