# NVIDIA/RULER

This repo contains the source code for RULER: What’s the Real Context Size of Your Long-Context Language Models?

Repository: https://github.com/NVIDIA/RULER
Canonical: https://ross.abutalabs.com/products/nvidia-ruler
Language: Python
License: Apache-2.0
License Family: permissive
Last push: 2026-07-22T17:32:53+00:00

## Health v2 (maintenance only)
Score: 66/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 93, release rhythm 35, longevity 62
- inputs: {"age_days": 876, "days_push": 42, "days_rel": null, "gap_med": null, "n_releases_24m": 0}
- flags: no_releases
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1609, forks 140 (observed 2026-08-28T04:05:10.868362+00:00)

## What it is
RULER is a benchmark framework from NVIDIA that generates synthetic test examples to measure the true effective context size of long-context language models. It provides configurable sequence lengths and 13 tasks across 4 categories, with reference results for many open-source and proprietary models.

## Use cases
- measure the real effective context length of an LLM
- benchmark long-context language models beyond simple needle-in-a-haystack tests
- evaluate how model performance degrades at 32k, 64k, and 128k tokens
- compare claimed vs actual context window sizes across models
- generate synthetic long-context evaluation datasets
- regression-test a fine-tuned model's long-context capabilities

## When to choose
- you need rigorous, configurable evaluation of long-context abilities
- you are comparing models' effective context sizes before deployment
- you want reproducible synthetic tasks instead of ad-hoc prompts

## When to avoid
- you need general-purpose LLM benchmarks unrelated to context length
- you lack GPU resources to run local model inference
- you only need simple needle-in-a-haystack checks

## Facets
- artifact type: library
- maturity: active
- function: benchmarking, llm-inference, data-generation, testing
- domain: large-language-models, machine-learning, artificial-intelligence, developer-tools
- platform: python
- tags: long-context, evaluation, synthetic-data, llm-benchmark, context-window, linux, gpu

## Member repositories
- NVIDIA/RULER (main) score 66

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:05:10.868362+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T03:51:15.636300+00:00, confidence not recorded.
  - readme: https://github.com/NVIDIA/RULER (fetched 2026-08-28T04:05:10.868362+00:00, sha 6eebf6f3f44e)
- Data as of 2026-08-30T08:39:29.467469+00:00.