|
--- |
|
language: |
|
- en |
|
license: mit |
|
--- |
|
# tiny-bert-ranker model card |
|
|
|
This model is a fine-tuned version of [prajjwal1/bert-tiny](https://web.archive.org/web/20240315094214/https://huggingface.co/prajjwal1/bert-tiny) |
|
as part of our submission to [ReNeuIR 2024](https://web.archive.org/web/20240704171521/https://reneuir.org/shared_task.html). |
|
|
|
## Model Details |
|
|
|
### Model Description |
|
|
|
<!-- Provide a longer summary of what this model is. --> |
|
|
|
The model is based on the pre-trained [prajjwal1/bert-tiny](https://huggingface.co/prajjwal1/bert-tiny). It is fine-tuned on a 1GB subset of data |
|
extracted from msmarco's [Train Triples Small](https://web.archive.org/web/20231209043304/https://microsoft.github.io/msmarco/Datasets.html). |
|
|
|
Tiny-bert-ranker is part of our investigation into the tradeoffs between efficiency and effectiveness in ranking models. |
|
This approach does not involve BM25 score injection or distillation. |
|
|
|
- **Developed by:** Team FSU at ReNeuIR 2024 |
|
- **Model type:** sequence-to-sequence model |
|
- **License:** mit |
|
- **Finetuned from model:** prajjwal1/bert-tiny |
|
|
