rotary-indictrans2-indic-en-mr-docs-v2

Rotary IndicTrans2 Marathi to English Fine-Tune

Continual fine-tune from anuj1541/rotary-indictrans2-indic-en-mr-docs on ~100k mix (full partial progress + long blocks).

Training mode: LoRA (lr=2e-05).

Load

from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
repo = "anuj1541/rotary-indictrans2-indic-en-mr-docs-v2"
tokenizer = AutoTokenizer.from_pretrained(repo, trust_remote_code=True)
model = AutoModelForSeq2SeqLM.from_pretrained(repo, trust_remote_code=True)

Training

• Direction: mar_Deva -> eng_Latn
• Epochs: 3
• Train pairs: 100,000
• Val pairs: 5,000
• Effective batch: 32
• LR: 2e-05
• Wall-clock: 1 day, 9:02:10 (33.04 h)
• Finished: 2026-09-16T20:47:54

Packaged files

pytorch_model.bin, config.json, generation_config.json, tokenization_indictrans.py, tokenizer_config.json, dict.SRC.json, dict.TGT.json, model.SRC, model.TGT, configuration_rotary_indictrans.py, modeling_rotary_indictrans.py

Eval (500 samples, MR→EN, seed=42)