Skip to content

搜索技能.../

Agent Skill Search Engine

搜索

搜索
分类
职业

关于

关于
隐私政策
服务条款

© 2026 Skills Pool. 保留所有权利。

Kv Lab Research | Skills Pool

技能档案

Kv Lab Research

Focused workflow for working inside kv-lab. Use when planning or implementing KV-cache compression, pruning, baseline harness work, model inspection, benchmarks, ablations, or artifact/reporting changes.

dheerajbatra0 星标2026年4月10日

职业
分类: 数据工程

技能内容

Use this skill when the task is specifically about kv-lab.

Goal

Keep Pi acting like a lean research harness, not a general-purpose coding assistant.

Default workflow

Inspect the relevant files first.
Anchor work to one of these buckets:
- baseline harness
- model/config inspection
- TurboQuant/offline quantization
- TriAttention/calibration/pruning
- benchmarks/artifacts/reporting
- documentation for experiment design/results
Make the smallest useful change.
Verify with the narrowest useful command.
Summarize what changed and what the next experiment should be.

Project-specific rules

Prefer reusable harness code over one-off scripts unless the task is clearly exploratory.
Prefer measurable outputs: JSON, JSONL, markdown summaries, benchmark artifacts.
Preserve comparability across runs.

相关技能

快速安装

Kv Lab Research

npx skillvault add dheerajbatra/dheerajbatra-kv-quant-lab-pi-skills-kv-lab-research-skill-md

下载 Skill 打开源码仓库

作者: dheerajbatra
星标: 0
更新时间: 2026年4月10日
职业

本页内容

Distinguish clearly between:

immediate implementation work,
deferred research ideas,
speculative future architecture.

Avoid adding unrelated Pi customizations unless they directly improve kv-lab research velocity.

For model work, start with the simplest supported local model before moving to harder hybrid/multimodal models.

For algorithm work, keep ablations visible:

MSE-only vs QJL
symmetric vs asymmetric K/V
baseline vs pruning vs quantization vs hybrid

Preferred file targets

Commonly relevant paths:

README.md
docs/
configs/
scripts/
src/kv_lab/
benchmarks/
artifacts/
.pi/

Verification guidance

After edits, prefer one of:

PYTHONPATH=src python3 -m compileall src tests scripts
PYTHONPATH=src python3 -m kv_lab.cli smoke
PYTHONPATH=src python3 -m kv_lab.cli profile <config>
a narrow run-baseline dry-run or mock-backend run

Output style

Be concise. State:

what changed,
what was verified,
what the next useful experiment is.

Default workflow

03Project-specific rules

04Preferred file targets

05Verification guidance

数据科学家

Clickhouse Io

ClickHouse database patterns, query optimization, analytics, and data engineering best practices for high-performance analytical workloads.

Clickhouse Io

고성능 분석 워크로드를 위한 ClickHouse 데이터베이스 패턴, 쿼리 최적화, 분석 및 데이터 엔지니어링 모범 사례.

Claude Devfleet

通过Claude DevFleet协调多智能体编码任务——规划项目、在隔离的工作树中并行调度智能体、监控进度并读取结构化报告。

Clickhouse Io

ClickHouse database patterns, query optimization, analytics, and data engineering best practices for high-performance analytical workloads.

Ai First Engineering

团队中人工智能代理生成大部分实施输出的工程运营模型。

Postgres Patterns

PostgreSQL database patterns for query optimization, schema design, indexing, and security. Based on Supabase best practices.

数据科学家