Linguistic Register Assessment on Synthetic VFR proxy corpus
6.1Lex F1 ScoreLlama 3.1 8B Instruct (Baseline)
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| Llama 3.1 8B Instruct (Baseline)Model Condition=unmodified, Sample Size (n)=100, Temperature=0.8, max_new_tokens=250, Backbone=Llama 3.1 8B Instruct2026.06 | 6.1 | 72.8 | 70.7 | 95.2 | 31.4 | — | — | — | — | — | |
| Llama 3.1 8B Instruct (Post-FT)Model Condition=LoRA fine-tune on 500-example VFR corpus, Sample Size (n)=100, Temperature=0.8, max_new_tokens=250, Backbone=Llama 3.1 8B Instruct2026.06 | 1 | 56.7 | 56.1 | 71.1 | 42 | — | — | — | — | — |