TartarusXXX/en-ur-marian-transformers

en-ur Marian model (Transformers)

Model summary

• Direction: en -> ur
• Architecture: Marian transformer
• Format: Hugging Face Transformers
• Primary converted checkpoint: best-bleu
• Training dataset selection: NLLB CCMatrix CCAligned OpenSubtitles Tanzil XLEnt \
• Validation set: openlanguagedata_flores_plus
• Test set recipe: openlanguagedata_flores_plus

Best validation metrics seen in training logs

• BLEU: 26.6596 at epoch 3 / update 152500
• CHRF: 54.7017 at epoch 3 / update 170000
• PERPLEXITY: 347.646 at epoch 1 / update 2500

Files

• .gitattributes
• config.json
• generation_config.json
• model.safetensors
• source.spm
• target.spm
• tokenizer_config.json
• vocab.json

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

repo_id = "YOUR_ORG/YOUR_MODEL"
tokenizer = AutoTokenizer.from_pretrained(repo_id)
model = AutoModelForSeq2SeqLM.from_pretrained(repo_id)

text = "Hello world!"
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs)
print(tokenizer.batch_decode(outputs, skip_special_tokens=True)[0])

Notes

• This model was converted from Marian .npz weights using the official Hugging Face Marian conversion module.
• Review dataset and license compatibility before redistributing the model publicly.