菜单
Guidance for querying ML model leaderboards and benchmarks (MTEB, HuggingFace, embedding benchmarks). This skill applies when tasks involve finding top-performing models on specific benchmarks, compar
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Guidance for building and training with the Caffe deep learning framework on CIFAR-10 dataset. This skill applies when tasks involve compiling Caffe from source, training convolutional neural networks on image classification datasets, or working with legacy deep learning frameworks that have compatibility issues with modern systems.
Guide for analyzing chess positions from images and determining optimal moves. This skill should be used when asked to find the best move, checkmate, or tactical solution from a chess board image. It provides structured approaches for image-based chess analysis, piece detection calibration, position validation, and move verification.
Guidance for building CompCert, a formally verified C compiler. This skill applies when tasks involve compiling CompCert from source, setting up Coq/OCaml environments with opam, or building software with strict proof assistant dependencies. Use for CompCert compilation, Coq-dependent project builds, or formal verification toolchain setup.
This skill provides guidance for cracking 7z archive password hashes. It should be used when tasks involve extracting hashes from password-protected 7z archives, selecting appropriate cracking tools, and recovering passwords through dictionary or brute-force attacks. Applicable to password recovery, security testing, and CTF challenges involving encrypted 7z files.
Guide for recovering data from SQLite Write-Ahead Log (WAL) files that may be corrupted, encrypted, or inaccessible through standard methods. This skill should be used when tasks involve SQLite database recovery, WAL file analysis, encrypted database files, or discrepancies between tool outputs and filesystem access.
Guidance for finding probability distributions that satisfy specific statistical constraints such as KL divergence targets, entropy requirements, or moment conditions. This skill should be used when tasks involve constructing discrete or continuous probability distributions with specified divergence measures, entropy values, or other distributional properties through numerical optimization.
| name | mteb-leaderboard |
| description | Guidance for querying ML model leaderboards and benchmarks (MTEB, HuggingFace, embedding benchmarks). This skill applies when tasks involve finding top-performing models on specific benchmarks, compar |
This skill provides guidance for accurately querying machine learning model leaderboards and benchmarks, particularly the Massive Text Embedding Benchmark (MTEB) and related embedding leaderboards.
Before searching for results, establish which sources contain authoritative, current data:
Primary Sources (prefer these):
mteb-leaderboard on HuggingFace Spaces)Secondary Sources (use with caution):
When a task specifies a time constraint (e.g., "as of August 2025"):
When web pages don't render properly (interactive charts, JavaScript-heavy pages):
Look for raw data endpoints:
/api/ or /data/ endpointsTry alternative access methods:
Search for data repositories:
site:github.com [leaderboard name] results jsonsite:huggingface.co [benchmark name] leaderboardDo not make assumptions about which models "count" on a leaderboard:
When primary access fails:
Academic papers have publication delays of 3-12 months. A paper published in June 2024 contains data from early 2024 at best. Never use paper results for questions about current standings.
Interactive leaderboards often don't render in simple web fetches. Always try:
Do not assume API models (OpenAI, Cohere, etc.) cannot be valid answers. Check the actual task requirements and leaderboard contents.
Before writing a final answer:
If a task asks about a specific date, ensure data sources reflect that timeframe. A 14-month gap between data and required date is unacceptable.
When searching for leaderboard information:
Start broad, then narrow:
[benchmark name] leaderboard 2025[benchmark name] top models currentsite:huggingface.co [benchmark name]Search for raw data:
[benchmark name] results github[benchmark name] json data[benchmark name] apiSearch for recent updates:
[benchmark name] new top model [current year][benchmark name] leaderboard updateAvoid repetitive similar queries - If a query pattern isn't working after 2-3 attempts, change the approach rather than making minor variations
Before submitting an answer, verify:
Overview
This skill provides practical guidance for querying machine learning model leaderboards and benchmarks such as MTEB, HuggingFace leaderboards, and embedding benchmarks. It focuses on finding top-performing models, comparing standings across sources, and handling time-sensitive queries while avoiding common pitfalls with outdated or interactive data sources.
How this skill works
First, identify authoritative sources (official leaderboard pages, GitHub data, or API/JSON endpoints). Then verify temporal alignment by checking last-updated timestamps and cross-referencing multiple sources. When pages are interactive or JavaScript-driven, locate raw data endpoints, CSV/JSON exports, or repository artifacts to extract canonical results. Finally, validate model eligibility and record any discrepancies.
When to use it
Finding the highest-ranked models on MTEB or other embedding leaderboards
Answering questions about leaderboard standings as of a specific date
Comparing model performance across multiple benchmarks
Retrieving live or recently-updated leaderboard data
Resolving conflicts between paper results and live leaderboards
Best practices
Prefer primary sources: official leaderboard pages, APIs, or GitHub-hosted raw data
Always check 'last updated' timestamps and be explicit about temporal gaps
Cross-reference at least 2–3 independent sources and prioritize the most recent authoritative one
Search for raw JSON/CSV endpoints or GitHub repositories before relying on rendered pages
Verify model eligibility and exact name formatting on the actual leaderboard
Example use cases
User asks: 'Which model ranked first on MTEB as of August 2025?' — locate the leaderboard snapshot or API data with that date and report the top model plus data source
Comparing two models across MTEB and a Scandinavian embedding benchmark to identify consistent top performers
Extracting CSV/JSON from a HuggingFace Spaces backend when the web UI is interactive or fails to render
Validating whether an API-based model qualifies on a benchmark that lists only open-source entries
FAQ
What if the leaderboard page is interactive and fetches data via JavaScript?
Look for underlying API endpoints, JSON/CSV exports, or the project's GitHub repo. Many Spaces or Gradio apps expose programmatic endpoints.
Can I use academic papers to answer current leaderboard queries?
Not reliably. Papers are often mon