Spaces:

Hugdrigo
/

evalmatrix-llm-benchmark

Running

App Files Files Community

evalmatrix-llm-benchmark / README.md

Hugdrigo

Upload 3 files

5d2ab24 verified 14 days ago

preview code

Raw History Blame Contribute Delete

2.68 kB

metadata

title: EvalMatrix | LLM Quality Evaluation Studio
emoji: 📊
colorFrom: blue
colorTo: green
sdk: static
app_file: index.html
pinned: false
license: mit
short_description: LLM quality and cost benchmarking.

EvalMatrix | Project #124

LLM Quality Evaluation Studio
LLM quality and cost benchmarking.

Compare synthetic language-model evaluations using quality, hallucination, latency, and price.

Features

Interactive overview with portfolio indicators, ranking charts, diagnostic scores, and explanations.
Four scenario presets with three editable, domain-specific numeric controls.
Fully editable synthetic records (create, update, archive and restore).
Dynamic calculations computed locally in JavaScript without paid services or network requests.
CSV results export and structured JSON backup/import with validation and file-size limits.
Local browser persistence, responsive mobile/desktop navigation, and light/dark theme.
Transparent methodology, clear synthetic-data labeling, and limitation statements.

Model methodology

Weighted quality combines accuracy, robustness, unsupported-response penalty, and blended input/output token costs.
Eligibility constraints filter models by p95 latency and output token price.
All measurements are synthetic and do not represent tested commercial models.

Recommended use: Review accuracy/cost tradeoffs and models excluded by hard constraints.

How to run

Hugging Face Spaces settings: SDK Static, application file index.html, visibility Public, license MIT.
Upload README.md, index.html and LICENSE directly to the Space root.
The index.html may also be opened locally in a current browser.

Important limits

The dataset, numeric inputs, names, economic estimates, probabilities and scenario outcomes are synthetic illustrations, not claims about actual persons, companies, locations, scientific findings, operational systems, market prices, or model performance. These are simplified prototyping and educational tools, not validated decision engines. The app does not fetch or send user data to external APIs. Browser data is local to the device/browser profile. Back up JSON before resetting or clearing your browser data. Do not upload sensitive records to public repos.

Quality & publication

Static SDK: sdk: static; entrypoint: app_file: index.html.
Supported Space card colors: colorFrom: blue, colorTo: green.
No external JavaScript/CSS/CDN dependencies.
Files at root: README.md, index.html, and LICENSE.