Skip to content
Back to skills

Retrieval Optimization

ASecurity

"Operate LimiX sample-retrieval ensemble inference and Optuna

  • 247 stars
  • 0 votes
  • 0 copies
  • 3 views
  • Added September 8, 2026
toolsapi

Works with

  • cli
  • api

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 8, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill retrieval-optimization --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Retrieval Optimization?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Retrieval Optimization
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-retrieval-optimization/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-retrieval-optimization)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: retrieval-optimization
description: "Operate LimiX sample-retrieval ensemble inference and Optuna
  search-space tuning."
disable-model-invocation: true
metadata:
  disco-role: operating
license: Apache 2.0
---

# Retrieval optimization

Use this sub-skill when you need to explain, preview, or tune the sample-retrieval ensemble path for a local LimiX checkpoint.

## What this covers
- sample-retrieval ensemble inference
- Optuna tuning over retrieval parameters
- attention-map generation for retrieval
- safe preview of the retrieval search space

## Route away
- Base predictor inputs, tensor shapes, and task-level inference: `../predictor-inference/SKILL.md`
- JSON schema, config generation, and config validation: `../configuration-preprocessing/SKILL.md`
- Benchmark CLI loops and `search_space_sample_num`: `../benchmark-cli/SKILL.md`

## Before you start
- Use a local LimiX checkpoint and a retrieval-enabled config list.
- Retrieval runs are CUDA/GPU-oriented; the shipped project notes that sample-retrieval ensemble inference is intended for very high-memory GPUs.
- Full checkpoint inference is not run here; use the bundled preview script first.

## Read first
- `references/api-reference.md`
- `references/workflows.md`
- `references/troubleshooting.md`
- `scripts/preview_retrieval_search_space.py`

## Safety rules
- Do not run Optuna tuning blindly at the default trial count on a large dataset.
- Do not pass a CPU device with `use_retrieval: true`.
- Do not assume the constructor's `attention_score` argument overrides the predictor's internal attention flow; the current search loop recomputes through the predictor path.

## Quick decision
- Need to inspect or validate ranges without running inference? Use `scripts/preview_retrieval_search_space.py`.
- Need to tune parameters for a local checkpoint? Follow `references/workflows.md`.
- Need to understand exact keys or signatures? Follow `references/api-reference.md`.
- Need a fix for OOM, missing attention, or config mismatch? Follow `references/troubleshooting.md`.

Files in this skill

  • SKILL.md2 KB
  • references/api-reference.md8.6 KB
  • references/troubleshooting.md4.2 KB
  • references/workflows.md5.2 KB
  • scripts/preview_retrieval_search_space.py19.3 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.