Hugging Face relaunches PapersWithCode with multi-metric leaderboards and 3,000+ model evals
Niels Rogge at Hugging Face relaunched PapersWithCode at paperswithcode.co with multi-metric leaderboard support — WER and RTFx in a single table — paper lineage banners tracking predecessors and follow-ups, AI-enriched task and method tags, and 3,000+ Transformers model evaluations pre-loaded at launch. The 3,000 evals cover every Transformers-supported model, so a developer can look up multi-metric benchmark numbers tied directly to the paper and GitHub repo that produced them — without cross-referencing separate leaderboard sites.
Source: huggingface.co ↗
leaderboards now support multiple metrics, see e.g., the Open ASR Leaderboard for automatic speech recognition, which supports both Word Error Rate (WER) and the Inverse Real-Time Factor (RTFx) metrics
Why this matters
- → Developers can now benchmark models against papers using unified multi-metric leaderboards instead of cross-referencing separate sites.
- → 3,000+ pre-loaded model evaluations let researchers instantly see how Transformers-supported models perform on published benchmarks.
- → Paper lineage tracking surfaces predecessors and follow-ups, clarifying the research trajectory in each domain.