zhusuan
Use ZhuSuan for Bayesian networks, variational inference, HMC/SG-MCMC sampling, and importance-sampling workflows on TensorFlow 1.x.
- License 许可
- MIT
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- zhusuan
- license
- MIT
- metadata
- { "disco-role": "operating" }
- description
- Use ZhuSuan for Bayesian networks, variational inference, HMC/SG-MCMC sampling, and importance-sampling workflows on TensorFlow 1.x.
- disable-model-invocation
- true
02 资源文件 9
- 参考文档references/api-reference.md3.0 KB
- 参考文档references/overview.md1.4 KB
- 参考文档references/repo-provenance.md734 B
- 参考文档references/repo-routing-metadata.json432 B
- 参考文档references/troubleshooting.md2.0 KB
- 参考文档references/workflows.md2.5 KB
- 脚本scripts/core_smoke.py1.2 KB
- 脚本scripts/data_helpers.py1.2 KB
- 脚本scripts/image_grid.py2.0 KB
03 SKILL.md 正文
ZhuSuan
ZhuSuan is a TensorFlow 1.x probabilistic-programming library for Bayesian deep
learning. Use this skill when the user asks about import zhusuan as zs,
Bayesian networks, latent-variable models, variational objectives, posterior
sampling, or the repo's example workflows.
Quick start
- Read
references/overview.mdfor the module map and installation note. - Read
references/workflows.mdfor the example-family map and dataset notes. - Read
references/api-reference.mdwhen you need exact signatures. - Read
references/troubleshooting.mdwhen install, TF1, dtype, or shape issues show up. - Run
scripts/core_smoke.pyafter installing the package if you want a fast import-and-objective sanity check.
Route to a sub-skill
sub-skills/modeling-primitives/SKILL.mdfor distributions, BayesianNet, MetaBayesianNet, observations, deterministic nodes, and node inspection.sub-skills/variational-inference/SKILL.mdfor ELBO, IWAE, KLPQ, importance-sampling likelihoods, normalizing flows, and VAE/BNN/SVGP-style training.sub-skills/mcmc-and-sampling/SKILL.mdfor HMC, SGLD, PSGLD, SGHMC, SGNHT, AIS, and chain diagnostics.
What this root skill is for
- Give the fastest route to the right workflow and reference.
- Provide the package-level install, import, and troubleshooting entry point.
- Keep the skill usable after the original checkout disappears.
What this root skill is not for
- It is not a full API manual; read the relevant reference file instead.
- It is not a training script collection; the example families are summarized in
references/workflows.mdand the sub-skills. - It is not a replacement for the sub-skills; those carry the detailed workflow guidance.
Practical reminders
- ZhuSuan's core API is graph-based and TF1-style. Sessions, placeholders, and variable scopes still matter.
- The verified inspection environment used TensorFlow 1.15.5 with SciPy and mock on Python 3.6.
- The example families often need external data or optional image helpers, so keep them as references unless the user explicitly wants a full run.
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).