# modelscope/evalscope

A streamlined and customizable framework for efficient large model (LLM, VLM, AIGC) evaluation and performance benchmarking.

Repository: https://github.com/modelscope/evalscope
Canonical: https://ross.abutalabs.com/products/evalscope
Homepage: https://evalscope.readthedocs.io/en/latest/
Language: Python
License: Apache-2.0
License Family: permissive
Topics: evaluation, llm, performance, rag, vlm
Last push: 2026-08-26T09:36:54+00:00

## Health v2 (maintenance only)
Score: 93/100 (v2, computed 2026-09-02T17:46:02.011165+00:00)
- activity 99, release rhythm 99, longevity 71
- inputs: {"age_days": 1000, "days_push": 7, "days_rel": 9, "gap_med": 13.0, "n_releases_24m": 49}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 3312, forks 467 (observed 2026-08-28T04:07:56.019488+00:00)

## What it is
EvalScope is a Python framework from the ModelScope community for evaluating large language models, vision-language models, embedding models, and AIGC models against built-in benchmarks like MMLU, C-Eval, and GSM8K. It also provides inference performance stress testing, agent-based evaluation with sandboxed tool use, multi-model arena battles, and a web dashboard for result visualization.

## Use cases
- benchmark an LLM on MMLU and GSM8K
- stress test inference performance of a model service
- compare two models head-to-head in an arena
- evaluate a vision-language model
- evaluate RAG pipelines with rerankers and embeddings
- run agentic benchmarks like SWE-bench in a sandbox
- visualize evaluation results in a dashboard

## When to choose
- you need a one-command evaluation pipeline for LLMs or VLMs via OpenAI-compatible APIs
- you want both quality benchmarks and inference performance metrics like TTFT and TPOT in one tool
- you need multi-backend support spanning OpenCompass, VLMEvalKit, and RAGEval
- you want pairwise model battles and interactive comparison reports

## When to avoid
- you only need simple unit testing of application code rather than model evaluation
- you require a fully managed cloud evaluation service with no local setup
- your models are not reachable via supported API or local inference backends

## Facets
- artifact type: framework
- maturity: active
- function: benchmarking, testing, machine-learning, llm-inference, rag, data-visualization
- domain: large-language-models, machine-learning, artificial-intelligence, developer-tools, performance
- platform: python, cli, cross-platform
- tags: llm-evaluation, vlm-evaluation, stress-testing, arena-mode, model-benchmarking, openai-api, agent-evaluation

## Member repositories
- modelscope/evalscope (main) score 93

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:07:56.019488+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 2026-08-29T18:42:24.081072+00:00, confidence not recorded.
  - readme: https://github.com/modelscope/evalscope (fetched 2026-08-28T04:07:56.019488+00:00, sha e0dba4034242)
  - registry_pypi: https://pypi.org/pypi/evalscope/json (fetched 2026-08-29T09:35:41.606743+00:00, sha 0022424af9c8)
- Data as of 2026-08-30T08:39:29.467469+00:00.