factorized_score_composition
Compose F-NPSE and PF-NPSE posterior scores from prior corrections and per-condition score estimates.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- factorized_score_composition
- description
- Compose F-NPSE and PF-NPSE posterior scores from prior corrections and per-condition score estimates.
02 资源文件 2
03 SKILL.md 正文
Factorized Score Composition
Use this skill when implementing or auditing the core F-NPSE/PF-NPSE inference-time score. It should be called before sampling so the recovery can prove that additive score composition, not duplicated harness logic, produced the score.
Inputs
- Current
thetavector. - A progress value
(T - t) / Tin[0, 1]. - Per-observation scores for F-NPSE, or per-subset scores for PF-NPSE.
- Prior score vector, typically
-thetaunder standard-normal reparameterization. - Condition count: number of observations for F-NPSE or number of subsets for PF-NPSE.
Outputs
- Composed score vector.
- Trace with score sum, prior correction, condition count, progress, and mode.
Workflow
- Validate all score vectors have the same dimensionality.
- Sum learned condition scores.
- Compute the correction
(1 - condition_count) * progress * prior_score. - Add the correction to the score sum.
- Return a trace so downstream recovery can verify the sign and count semantics.
Validation
Run python tests/test_factorized_score_composition.py or the Distiller skill-tree validator.
Limitations
This skill composes supplied score estimates. It does not train the score network and does not sample posterior particles by itself.
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).