Patent Claim Generation on USPTO HUPD 2017 (test)
49.7BLEUAdaptive Multi-Stage Framework
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| Adaptive Multi-Stage FrameworkFine-tuning=Domain-adaptive LoRA, Chunking=Adaptive chunking mechanism, Attention=multi-head, Embedding dimension=5122026.01 | 49.7 | 52.8 | 91.2 | 8.4 | 8.6 | 8.2 | |
| PatClaimEval2026.01 | 44.3 | 48.1 | 88.9 | 7.8 | 8 | 7.6 | |
| GPT-4oModel type=Large Language Model2026.01 | 42.1 | 45.2 | 87.8 | 7.6 | 8.3 | 7.4 | |
| Llama-3.1-8BFine-tuning=Standard LoRA, Base Model=Llama-3.1-8B2026.01 | 38.9 | 41.7 | 84.1 | 7.1 | 7.8 | 6.9 |