Model Card for passage-ranker.nectarine
This model is a passage ranker developed by Sinequa. It produces a relevance score given a query-passage pair and is used to order search results.
Model name: passage-ranker.nectarine
Supported Languages
The model was trained and tested in the following languages:
- English
- French
- German
- Spanish
- Italian
- Dutch
- Japanese
- Portuguese
- Chinese (simplified)
- Polish
- Arabic
- Korean
Besides the aforementioned languages, basic support can be expected for additional 93 languages that were used during the pretraining of the base model (see
list of languages).
Scores
| Metric | Value |
|---|
| English Relevance (NDCG@10) | 0.455 |
| Arabic Relevance (NDCG@10) | 0.250 |
| Korean Relevance (NDCG@10) | 0.232 |
Note that the relevance score is computed as an average over several retrieval datasets (see
details below).
Inference Times
| GPU | Quantization type | Batch size 1 | Batch size 32 |
|---|
| NVIDIA A10 | FP16 | 2 ms | 28 ms |
| NVIDIA A10 | FP32 | 4 ms | 82 ms |
| NVIDIA T4 | FP16 | 3 ms | 65 ms |
| NVIDIA T4 | FP32 | 14 ms | 369 ms |
| NVIDIA L4 | FP16 | 3 ms | 38 ms |
| NVIDIA L4 | FP32 | 5 ms | 123 ms |
GPU Memory usage
| Quantization type | Memory |
|---|
| FP16 | 850 MiB |
| FP32 | 1200 MiB |
Note that GPU memory usage only includes how much GPU memory the actual model consumes on an NVIDIA T4 GPU with a batch
size of 32. It does not include the fix amount of memory that is consumed by the ONNX Runtime upon initialization which
can be around 0.5 to 1 GiB depending on the used GPU.
Requirements
- Minimal Sinequa version: 11.10.0
- CUDA compute capability: above 7.5
Model Details
Overview
- Number of parameters: 167 million
- Base language model: Multilingual BERT-Base
- Insensitive to casing and accents
- Training procedure: MonoBERT
Training Data
Evaluation Metrics
English
To determine the relevance score, we averaged the results that we obtained when evaluating on the datasets of the
BEIR benchmark. Note that all these datasets are in English.
| Dataset | NDCG@10 |
|---|
| Average | 0.455 |
| |
| Arguana | 0.501 |
| CLIMATE-FEVER | 0.200 |
| DBPedia Entity | 0.353 |
| FEVER | 0.723 |
| FiQA-2018 | 0.299 |
| HotpotQA | 0.657 |
| MS MARCO | 0.406 |
| NFCorpus | 0.299 |
| NQ | 0.449 |
| Quora | 0.751 |
| SCIDOCS | 0.136 |
| SciFact | 0.605 |
| TREC-COVID | 0.694 |
| Webis-Touche-2020 | 0.296 |
Arabic
This model has arabic capacities, that are being evaluated over a home made translation of Msmarco with BM25 as the first stage retrieval.
| Dataset | NDCG@10 |
|---|
| msmarco-ar | 0.250 |
Korean
This model has korean capacities, that are being evaluated over a home made translation of Msmarco with BM25 as the first stage retrieval.
| Dataset | NDCG@10 |
|---|
| msmarco-ko | 0.232 |
Other languages
We evaluated the model on the datasets of the
MIRACL benchmark to test its
multilingual capacities. Note that not all training languages are part of the benchmark, so we only report the metrics
for the existing languages.
| Language | NDCG@10 |
|---|
| French | 0.390 |
| German | 0.371 |
| Spanish | 0.447 |
| Japanese | 0.488 |
| Chinese (simplified) | 0.429 |
