Machine Translation on IN22 Conversational 1.0
18.95BLEUIndicTrans2 en indic 200 M (Original)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| IndicTrans2 en indic 200 M (Original)Model Parameters=200 M2025.12 | 18.95 | 0.8209 | 0.8117 | 51.05 | 47.74 | 0.9051 | 0.7451 | |
| Finetuned (with and w/o punct) (2x)Training Data=with and without punctuation, Data Scale=2x2025.12 | 17.93 | 0.8106 | 0.8065 | 50.45 | 47.33 | 0.8962 | 0.7408 | |
| Finetuned (alternate with and w/o punct) (x)Training Mode=alternate punctuation, Data Scale=1x2025.12 | 17.87 | 0.8122 | 0.8082 | 50.34 | 47.26 | 0.8983 | 0.7411 | |
| Fine-tuned t5-base + Original ModelRestoration Backbone=t5-base, Translation Backbone=IndicTrans22025.12 | 17.82 | 0.8121 | 0.8077 | 50.24 | 47.17 | 0.8982 | 0.7404 | |
| Finetuned (w/o punct) (x)Training Data=without punctuation, Data Scale=1x2025.12 | 16.67 | 0.8034 | 0.7995 | 49.68 | 46.72 | 0.8927 | 0.7354 | |
| Finetuned (w/ punct) (x)Training Data=with punctuation, Data Scale=1x2025.12 | 16.08 | 0.7963 | 0.792 | 49.77 | 46.65 | 0.8869 | 0.73 |