README.md

---
language:
- vi
- ede
tags:
- cross-lingual-retrieval
- morpheme-tokenizer
- vanilla-transformer
- EViRAL
---

# Vanilla Transformer + Morpheme Tokenizer — EViRAL

Task: Ede query → Vietnamese passage retrieval
Config: 6 layers / hidden 512 / 8 heads / FFN 2048
Tokenizer: corpus-driven morpheme segmentation + Ede-only synonym buffer (Vi as pivot)

## Checkpoints
| file | description |
|---|---|
| mlm.pt | MLM pre-trained encoder |
| align.pt | cross-lingual aligned encoder |
| finetune.pt | contrastive fine-tuned encoder (best val) |
| vocab.json | morpheme vocab (token → id) |

## Vocab size
57023