GLiNER2 unifies NER, relation extraction and classification in one model under 100ms locally
philipvollet · x · 2026-09-25
GLiNER2, an open-source schema-driven information extraction framework (2.2k stars on GitHub), now offers a unified model family:
- A single forward pass handles JEVS-style classification, NER, relation extraction, structured record extraction, and span attributes — all under 100ms on local inference.
- Two architectures are supported: span (fixed-width span grid, compatible with legacy checkpoints like GLiGuard and PII fine-tunes) and boundary (GLiNER2.5, sparse start/end pairing for arbitrary-length spans).
- AutoExtractor.frompretrained() dispatches by the saved architecture field; note that GLiNER2.frompretrained() remains span-only and won't load GLiNER2.5 boundary checkpoints.
More from coding & agent
- Runway MCP Lets Any AI Agent Generate Video Directly — tlakomy · 2026-09-25
- Hexis: open-source Postgres-native cognitive architecture giving LLMs memory and identity — GeretsOliver · 2026-09-25
- AgentArk, distilling multi-agent debate into a single LLM, accepted to NeurIPS 2026 — jindong_wang92 · 2026-09-25
- Ben Tossell Built an 87-Device Retro Timeline Site in a Morning with Codex and Subagents — Ben's Bites · 2026-09-25
- video-shotcraft author: Opus 5.5 makes a full code-generated video with music in one prompt — caolonghua · 2026-09-25
- Stability Matrix hang on Linux traced to 143 Windows fonts breaking Fontconfig — Signal_Care6558 · 2026-09-25