Text2Cypher on Text2Cypher (Avg English, Spanish, Turkish)
85ROUGE-LJoint Multilang FT
Evaluation results
| Method | Links | |
|---|---|---|
| Joint Multilang FTBackbone=Meta-Llama-3.1-8B-Instruct, Train samples=~36K2026.01 | 85 | |
| EN-Only FTBackbone=Meta-Llama-3.1-8B-Instruct, Train samples=~20.5K2026.01 | 79 | |
| Fusion MLPBackbone=Meta-Llama-3.1-8B-Instruct, Gating=Dynamic, Tuning samples=7,5002026.01 | 79 | |
| Linear MergeBackbone=Meta-Llama-3.1-8B-Instruct, Merging weights=Uniform (0.333)2026.01 | 75 | |
| BaseBackbone=Meta-Llama-3.1-8B-Instruct2026.01 | 60 |