Update model card
#2
by
boris-f
- opened
README.md
CHANGED
|
@@ -1,30 +1,63 @@
|
|
| 1 |
---
|
| 2 |
language:
|
|
|
|
| 3 |
- ja
|
|
|
|
| 4 |
---
|
| 5 |
|
| 6 |
-
# Model Card for answer-finder-v1-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
-
|
| 9 |
|
| 10 |
-
|
| 11 |
|
| 12 |
- Japanese
|
| 13 |
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
|:-----------------|:---:|
|
| 17 |
-
|__Japanese__ | <span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:#ffd21e;">★</span><span style="font-size:200%;color:black;">★</span> |
|
| 18 |
-
||
|
| 19 |
|
| 20 |
-
|
| 21 |
-
|
| 22 |
-
|
| 23 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 24 |
|
|
|
|
|
|
|
|
|
|
| 25 |
|
| 26 |
-
|
| 27 |
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
| [JSQuAD](https://github.com/yahoojapan/JGLUE) | [paper](https://aclanthology.org/2022.lrec-1.317.pdf) |
|
|
|
|
| 1 |
---
|
| 2 |
language:
|
| 3 |
+
|
| 4 |
- ja
|
| 5 |
+
|
| 6 |
---
|
| 7 |
|
| 8 |
+
# Model Card for answer-finder-v1-L-ja
|
| 9 |
+
|
| 10 |
+
This model is a question answering model developed by Sinequa. It produces two lists of logit scores corresponding to
|
| 11 |
+
the start token and end token of an answer.
|
| 12 |
+
|
| 13 |
+
Model name: `answer-finder-v1-L-ja`
|
| 14 |
|
| 15 |
+
## Supported Languages
|
| 16 |
|
| 17 |
+
The model was trained and tested in the following languages:
|
| 18 |
|
| 19 |
- Japanese
|
| 20 |
|
| 21 |
+
Besides the aforementioned languages, basic support can be expected for the 104 languages that were used during the
|
| 22 |
+
pretraining of the base model (See [original repository](https://github.com/google-research/bert)).
|
|
|
|
|
|
|
|
|
|
| 23 |
|
| 24 |
+
## Scores
|
| 25 |
+
|
| 26 |
+
| Metric | Value |
|
| 27 |
+
|:--------------------------------------------------------------|-------:|
|
| 28 |
+
| F1 Score on JSQuAD with Hugging Face evaluation pipeline | 92.1 |
|
| 29 |
+
| F1 Score on JSQuAD with Haystack evaluation pipeline | 91.5 |
|
| 30 |
+
|
| 31 |
+
## Inference Time
|
| 32 |
+
|
| 33 |
+
| GPU | Batch size 1 | Batch size 32 |
|
| 34 |
+
|:--------------------------------------------------------------|---------------:|---------------:|
|
| 35 |
+
| NVIDIA A10 | 4 ms | 84 ms |
|
| 36 |
+
| NVIDIA T4 | 15 ms | 361 ms |
|
| 37 |
+
|
| 38 |
+
The inference times only measure the time the model takes to process a single batch, it does not include pre- or
|
| 39 |
+
post-processing steps like the tokenization.
|
| 40 |
+
|
| 41 |
+
**Note that the Answer Finder models are only used at query time.**
|
| 42 |
+
|
| 43 |
+
## Requirements
|
| 44 |
+
|
| 45 |
+
- Minimal Sinequa version: 11.10.0
|
| 46 |
+
- GPU memory usage: TODO
|
| 47 |
+
|
| 48 |
+
Note that GPU memory usage only includes how much GPU memory the actual model consumes on an NVIDIA T4 GPU with a batch
|
| 49 |
+
size of 32. It does not include the fix amount of memory that is consumed by the ONNX Runtime upon initialization which
|
| 50 |
+
can be around 0.5 to 1 GiB depending on the used GPU.
|
| 51 |
+
|
| 52 |
+
## Model Details
|
| 53 |
+
|
| 54 |
+
### Overview
|
| 55 |
|
| 56 |
+
- Number of parameters: 110 million
|
| 57 |
+
- Base language model: [bert-base-multilingual-cased](https://huggingface.co/bert-base-multilingual-cased)
|
| 58 |
+
- Sensitive to casing and accents
|
| 59 |
|
| 60 |
+
### Training Data
|
| 61 |
|
| 62 |
+
- [JSQuAD](https://github.com/yahoojapan/JGLUE) see [Paper](https://aclanthology.org/2022.lrec-1.317.pdf)
|
| 63 |
+
- Japanese translation of SQuAD v2 "impossible" query-passage pairs
|
|
|