As part of ongoing efforts to make Information Retrieval (IR) more inclusive, this model addresses the needs of low-resource languages, focusing specifically on Urdu.
We created this model by translating the MS-Marco dataset into Roman-Urdu using the IndicTrans2 model.
To establish baseline performance, we initially tested for zero-shot learning for IR in Roman-Urdu using the unicamp-dl/mt5-base-mmarco-v2 model
and then applied fine-tuning with the mMARCO multilingual IR methodology on the translated dataset, resulting in State-Of-The-Art results for urdu IR
Although this model performs well and is state-of-the-art for now. But still this model is finetuned on mmarco model and a translated dataset(which was created using indicTrans2 model). Hence the limitations of those apply here too.
The evaluation was done using the scripts in the pygaggle library. Specifically these files:
evaluate_monot5_reranker.py
ms_marco_eval.py
1{
2 "_name_or_path": "unicamp-dl/mt5-base-mmarco-v2",
3 "architectures": ["MT5ForConditionalGeneration"],
4 "d_model": 768,
5 "num_heads": 12,
6 "num_layers": 12,
7 "dropout_rate": 0.1,
8 "vocab_size": 250112,
9 "model_type": "mt5",
10 "transformers_version": "4.45.2"
11}
For more details on how to customize the decoding parameters (such as max_length, num_beams, and early_stopping), refer to the Hugging Face documentation.
