stochastic_vi_chain
Run bounded constant-step noisy stochastic optimization chains that model stochastic VI iterates as a Markov process around an optimum.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- stochastic_vi_chain
- description
- Run bounded constant-step noisy stochastic optimization chains that model stochastic VI iterates as a Markov process around an optimum.
02 资源文件 3
- 脚本scripts/robust_svi.py3.9 KB
- 脚本scripts/simulate_chains.py3.6 KB
- 脚本tests/test_stochastic_vi_chain.py725 B
03 SKILL.md 正文
Stochastic VI Chain
Use this skill when a recovery or implementation needs deterministic evidence for the paper's view of stochastic VI optimization as a Markov chain. Do not use it to claim full probabilistic-programming VI training unless the gradient oracle is connected to a real model.
Inputs
chains: positive number of independent optimization runs.iterations: positive number of stochastic update steps.dimensions: positive variational parameter dimension.learning_rate,noise_scale,seed, and optional optimum vector.
Outputs
- JSON containing
iterates,gradients,initial_parameters,optimum, and run metadata. - Chain-major arrays:
chains x iterations x dimensions.
Workflow
- Choose or supply a known optimum.
- Initialize chains away from the optimum with deterministic seed control.
- At each step, evaluate a negative-quadratic gradient plus zero-mean Gaussian noise.
- Apply a constant-step stochastic gradient update and record the trace.
- Use the trace for downstream Rhat, ESS, MCSE, and iterate-averaging checks.
Validation
Run:
python3 scripts/simulate_chains.py --chains 4 --iterations 120 --dimensions 5 --output /tmp/stochastic_vi_chains.json
python3 tests/test_simulate_chains.py
Limitations
The bundled simulator is a reduced proxy for optimizer dynamics. It preserves stochastic iterate behavior and trainable-parameter updates but does not replace a full model-specific ELBO gradient.
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).