denoising_score_training
Train and validate conditional denoising score models for F-NPSE or PF-NPSE simulation-based inference recoveries.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- denoising_score_training
- description
- Train and validate conditional denoising score models for F-NPSE or PF-NPSE simulation-based inference recoveries.
02 资源文件 2
- 脚本scripts/score_training.py6.0 KB
- 脚本tests/test_score_training.py888 B
03 SKILL.md 正文
Denoising Score Training
Use this skill when a recovery or implementation needs the score-matching part of Compositional Score Modeling for Simulation-Based Inference. It is appropriate for NPSE, F-NPSE, and PF-NPSE score learners that receive simulator-generated (theta, condition) pairs and predict the Gaussian denoising score at a selected diffusion level.
Do not use this skill to combine multiple observations or to sample from the posterior. Score composition belongs to factorized_score_composition; posterior sampling belongs to annealed_langevin_sampler.
Inputs
- Clean simulator parameter vectors
theta_clean. - Conditioning values: one observation for F-NPSE, or a small observation group for PF-NPSE.
- Diffusion coefficient
gammawith0 < gamma < 1. - A parameterized score predictor. The bundled script provides a small linear conditional predictor for bounded recovery experiments.
Outputs
- Gaussian denoising target scores.
- Mean squared denoising score loss.
- Optimizer trace with
loss_before,loss_after,params_before, andparams_after. - A predictor callable or serialized parameter dictionary usable by composition and recovery scripts.
Workflow
- Generate noisy parameters with
theta_t = sqrt(gamma) * theta_clean + sqrt(1 - gamma) * epsilon. - Compute the target score
(sqrt(gamma) * theta_clean - theta_t) / (1 - gamma). - Evaluate the conditional score predictor on
theta_t,gamma, and the condition. - Minimize the squared error against the target.
- Record parameter movement and before/after loss whenever this is used as reduced recovery evidence.
Validation
Run the deterministic test suite:
python /share/project/yuyang/workspace/Paper2Skills/Distiller/skills/module-to-skill/scripts/validate_skill_tree.py <skill_dir> --run-tests
The tests use a Gaussian fixture and verify finite denoising targets, decreasing or non-increasing loss after a small gradient step, and changed trainable parameters.
Limitations
The bundled linear predictor is a recovery aid, not the paper's full neural architecture. A full reproduction should replace it with a neural score network while preserving the same denoising target and trace contract.
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).