This model is a fine-tuned version of the Helsinki-NLP OPUS-MT model for multiple language pairs. It has been fine-tuned on the Tatoeba dataset for the following language pairs:
English to Marathi (en-mr)
Esperanto to Dutch (eo-nl)
Spanish to Portuguese (es-pt)
French to Russian (fr-ru)
Spanish to Galician (es-gl)
The model supports sequence-to-sequence translation and has been optimized for performance using FP16 quantization.
Model Details
Base Model: Helsinki-NLP/opus-mt-en-roa
Training Dataset: Tatoeba dataset
Fine-tuned Language Pairs: en-mr, eo-nl, es-pt, fr-ru, es-gl
Evaluation Metric: BLEU Score (using sacreBLEU)
Training Framework: Hugging Face Transformers
Training Configuration
Optimizer: AdamW
Learning Rate: 2e-5
Batch Size: 16 (per device)
Weight Decay: 0.01
Epochs: 3
Precision: FP32 (initial training), converted to FP16 for inference
Quantization and FP16 Conversion
To improve inference efficiency, models were converted to FP16:
Inference Example
python
from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
import torch
model = AutoModelForSeq2SeqLM.from_pretrained("fine_tuned_models_fp16/en-mr/final/", torch_dtype=torch.float16).to("cuda")
tokenizer = AutoTokenizer.from_pretrained("fine_tuned_models_fp16/en-mr/final/")
inputs = tokenizer("Hello, how are you?", return_tensors="pt").to("cuda")
outputs = model.generate(**inputs)
print(tokenizer.decode(outputs[0], skip_special_tokens=True))
Usage
The models can be used for translation tasks in various NLP applications, including chatbots, document translation, and real-time communication.
Limitations
May not generalize well for domain-specific text.
FP16 quantization may lead to minor loss in precision.
Translation accuracy depends on the dataset quality.
Citation
If you use this model, please cite the original OPUS-MT paper and acknowledge the fine-tuning process conducted using the Tatoeba dataset.
