score_sde_kernels
Define Score SDE VE, VP, and sub-VP kernels with reverse and probability-flow drift checks.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- score_sde_kernels
- description
- Define Score SDE VE, VP, and sub-VP kernels with reverse and probability-flow drift checks.
02 资源文件 2
- 脚本scripts/sde_kernels.py6.0 KB
- 脚本tests/test_sde_kernels.py1.7 KB
03 SKILL.md 正文
Score SDE Kernels
Use this skill when you need the forward perturbation kernels from "Score-Based Generative Modeling through Stochastic Differential Equations" or when a recovery experiment needs reverse-time SDE or probability-flow drift algebra without the original repository.
Do not use this skill as a replacement for a trained score network, FID evaluation, or a full image-generation pipeline. It supplies the deterministic SDE math that other skills can consume.
Inputs
family: one ofve,vp, orsubvp.- Schedule parameters:
sigma_minandsigma_maxfor VE;beta_minandbeta_maxfor VP and sub-VP. - State values as scalars or lists of floats.
- Time
tin[0, 1]. - Optional score callback
score_fn(x, t)for reverse-time or probability-flow drifts.
Outputs
- Forward drift and diffusion.
- Gaussian marginal mean and standard deviation.
- Prior samples and prior log densities.
- Reverse-time SDE drift
f - g^2 score. - Probability-flow ODE drift
f - 0.5 g^2 score.
Workflow
- Instantiate
SDEKernelfromscripts/sde_kernels.pywith the required family and schedule. - Use
sde(x, t)for forward drift and diffusion. - Use
marginal_prob(x0, t)to perturb data and construct score-matching targets. - Use
reverse_drift(x, t, score_fn, probability_flow=False)for reverse SDE drift. - Use
reverse_drift(..., probability_flow=True)for the probability-flow ODE drift. - Save numerical checks in recovery logs when accepting a proxy experiment.
Validation
Run:
python scripts/sde_kernels.py --self-test
python tests/test_sde_kernels.py
The tests verify VP/sub-VP marginal bounds, VE prior log probability, and the half-correction identity for probability flow.
Limitations
This skill intentionally uses standard-library math and one-dimensional or vector fixtures. It preserves the paper's SDE contracts, but it does not implement neural architectures, dataset pipelines, or high-resolution image sampling.
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).