importance_rank_allocator
Compute AdaLoRA sensitivity-uncertainty triplet scores and globally mask singular values to a target rank budget.
- License 许可
- Apache-2.0
- 最近更新
- 2026-09-03
- Stars
- ★ 351
- Forks
- 27
01 前言 Front Matter
- name
- importance_rank_allocator
- description
- Compute AdaLoRA sensitivity-uncertainty triplet scores and globally mask singular values to a target rank budget.
02 资源文件 2
- 脚本scripts/rank_allocator.py3.1 KB
- 脚本tests/test_rank_allocator.py646 B
03 SKILL.md 正文
Importance-Aware Rank Allocator
Use this skill when a recovery or implementation needs AdaLoRA's adaptive rank allocation rather than fixed LoRA ranks.
Inputs
- A list of adapted matrices with
A,E,Bparameters and matching gradients. - Previous EMA state for sensitivity and uncertainty, or empty state for the first step.
- Hyperparameters
beta1,beta2, and target global rank.
Outputs
- Updated EMA state.
- Per-triplet importance scores.
- Mask threshold and masked singular values.
- Rank pattern for every adapted matrix.
Workflow
- Compute instantaneous sensitivity
abs(parameter * gradient)for every AdaLoRA parameter. - Update smoothed sensitivity and uncertainty EMAs.
- Score each entry with
smoothed_sensitivity * uncertainty. - Combine each triplet as singular-value score plus mean scores from the corresponding
Arow andBcolumn. - Retain the top target-rank triplets across all matrices and set other singular values to zero.
- Log rank pattern and threshold so downstream recovery can verify budget compliance.
Validation
Run python scripts/rank_allocator.py --self-test or the provided tests.
Limitations
This script is deterministic and standard-library only. It is intended as executable mechanism evidence and a reference contract for tensor implementations.
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).