generative_latent_mir
Score latent replay candidates with KL drift, entropy confidence penalties, and diversity filtering for MIR.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- generative_latent_mir
- description
- Score latent replay candidates with KL drift, entropy confidence penalties, and diversity filtering for MIR.
02 资源文件 2
- 脚本scripts/latent_mir.py2.5 KB
- 脚本tests/test_latent_mir.py1.3 KB
03 SKILL.md 正文
Generative and Latent MIR
Use this skill when MIR must operate over generated samples, compressed memories, or a deterministic latent proxy rather than raw replay examples. It captures the paper's generative MIR mechanism without requiring a VAE for small validation tests.
Inputs
- Latent candidate ids and vectors.
- Pre-update and virtual-update class probability distributions for each candidate.
- Entropy weight, diversity distance threshold, and replay budget.
- Optional nearest-neighbor mapping from latent points to stored compressed examples.
Outputs
- KL, entropy, and combined score per candidate.
- A diversity-filtered ranked selection.
- Mechanism diagnostics showing KL drift and entropy confidence were evaluated.
Workflow
- Normalize candidate probability vectors with small numerical smoothing.
- Compute
KL(y_pre || y_hat)for prediction drift after the virtual classifier update. - Compute entropy of
y_preand subtractentropy_weight * entropy. - Sort candidates by descending score.
- Apply greedy diversity filtering by Euclidean distance in latent space.
- Return selected candidates and diagnostics for recovery logging.
Validation
Run the included tests or validate_skill_tree.py --run-tests. Tests cover KL ranking, entropy penalties, smoothing, and diversity filtering using standard-library numeric fixtures.
Limitations
This helper validates the latent MIR objective and selection contract. It does not train an encoder, decoder, or VAE; a full neural implementation should provide probability vectors and latent candidates from the model stack while preserving the same output schema.
Validation note: the diversity smoke test intentionally verifies that near-duplicate latent candidates can be suppressed while a distant interfered candidate remains selected.
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).