bbvi_variance_reduction
Compute BBVI Rao-Blackwellized and score-function control-variate gradient estimators with variance diagnostics.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- bbvi_variance_reduction
- description
- Compute BBVI Rao-Blackwellized and score-function control-variate gradient estimators with variance diagnostics.
02 资源文件 2
03 SKILL.md 正文
BBVI Variance Reduction
When To Use
Use this skill after forming BBVI score-function terms when you can identify local model terms for a variational factor or want to apply the paper's score-function control variate. It preserves black-box behavior because it requires only local log-density values and score functions.
Do not use it to update parameters or to claim variance reduction without numeric diagnostics.
Inputs
score: sample-by-parameter score values for one variational factor.local_signal: per-sample local learning signal for that factor.full_signal: optional per-sample full learning signal for naive comparison.
Outputs
rao_blackwell_termsandrao_blackwell_estimate.control_variate_scaleandcontrol_variate_estimate.variance: naive, Rao-Blackwellized, and control-variate empirical variances.variance_reduction_ratio: naive variance divided by control-variate variance when available.
Workflow
- Normalize scores and validate sample counts.
- Compute Rao-Blackwellized local terms as
score * local_signal. - Estimate the control-variate scale by aggregated covariance divided by aggregated score variance.
- Compute corrected terms as local terms minus
a_hat * score. - Report variance diagnostics over per-sample term norms.
Validation
Run:
python tests/test_variance_reduction.py
Limitations
The caller must supply valid local terms. This skill does not infer a Markov blanket from an arbitrary model graph.
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).