stochastic_interpolant_sampler
Integrate bounded stochastic-interpolant probability-flow ODE and simple diffusion sampler updates from learned velocity and score fields.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- stochastic_interpolant_sampler
- description
- Integrate bounded stochastic-interpolant probability-flow ODE and simple diffusion sampler updates from learned velocity and score fields.
02 资源文件 2
- 脚本scripts/sampler.py2.3 KB
- 脚本tests/test_sampler.py775 B
03 SKILL.md 正文
Stochastic Interpolant Sampler
Use this skill after velocity and score or denoiser fields have been defined. It provides bounded deterministic ODE and seeded Euler-Maruyama style SDE updates for recovery experiments.
Inputs
- Initial scalar samples or sample summaries.
- A velocity function
b(t, x). - Optional score function
s(t, x). - A scalar diffusion coefficient
epsilonand step count. - A deterministic seed for stochastic updates.
Outputs
- Final generated samples.
- Trajectory summaries for initial and final states.
- Solver metadata tying the result to ODE or SDE dynamics.
Workflow
- Use ODE mode for the main reduced recovery unless stochastic diffusion behavior is being ablated.
- Apply Euler updates on a bounded uniform grid.
- For SDE mode, use drift
b + epsilon*sand seeded Gaussian increments. - Treat
epsilon=0as deterministic and verify it matches ODE drift-only behavior. - Log step count and sample summaries for validation.
Validation
Run:
python scripts/sampler.py --demo
python tests/test_sampler.py
Limitations
The script is intentionally small and deterministic. It is not a substitute for the paper's adaptive Dormand-Prince or Heun samplers, but it preserves the probability-flow and tunable-diffusion contracts for bounded recovery.
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).