GLiNER (Generalist Lightweight NER)
Bidirectional Transformer Encoder with Span Representations for Zero-Shot Open Entity Extraction
#NER#Zero-Shot#Span Representations#Bi-Encoder#GLiNER#Entity Extraction
Choose Presentation Mode:
STAGE 1 / 7— Anti-Pattern
Section 1: The LLM Anti-Pattern vs Right-Sized Model
The Naive Generative LLM Approach:
Sending 5-page legal contracts to a 70B parameter LLM with a 500-token prompt: "Extract all companies, dates, and contract values in valid JSON with exact offsets".
Why It Fails in Production:
LLMs hallucinate entity keys, fail on exact character offset spans, mangle JSON schemas on edge cases, stream tokens at 50 tokens/sec, and cost $0.03 per page.
Targeted Algorithm (GLiNER (Generalist Lightweight NER))
Latency:15ms (CPU/ONNX)
Cost / 1M Ops:$0.00
Determinism:100% Deterministic Character Spans
Generative LLM Alternative
Latency:3,500ms
Cost / 1M Ops:$12,000
Determinism:Prone to Hallucinated JSON & Missed Spans