Update README.md
Browse files
README.md
CHANGED
@@ -39,13 +39,13 @@ Evaluation results
|
|
39 |
Benchmark
|
40 |
---------
|
41 |
|
42 |
-
We compare the [DistilCamemBERT](https://huggingface.co/cmarkea/distilcamembert-base) model with 2 other modelizations working on french language. The first one [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) is based on well named [CamemBERT](https://huggingface.co/camembert-base), the french RoBERTa model and the second one [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) based on [mDeBERTav3](https://huggingface.co/microsoft/mdeberta-v3-base) a multilingual model. To compare the performances the
|
43 |
|
44 |
-
| **NLI** | **time (ms)** | **MCC (x100)** |
|
45 |
-
| :--------------: | :-----------: | :------------: |
|
46 |
-
| [cmarkea/distilcamembert-base-nli](https://huggingface.co/cmarkea/distilcamembert-base-nli) | **51.35** | 66.24
|
47 |
-
| [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) | 105.0 | 72.67 |
|
48 |
-
| [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) | 299.18 | **75.15**
|
49 |
|
50 |
Zero-shot classification
|
51 |
------------------------
|
@@ -55,19 +55,19 @@ $$P(hypothesis=c|premise)=\frac{e^{P(premise=entailment\vert hypothesis\; c)}}{\
|
|
55 |
|
56 |
For this part, we use 2 datasets, the first one: [allocine](https://huggingface.co/datasets/allocine) used to train the sentiment analysis models. The dataset is composed of 2 classes: "positif" and "négatif" appreciation of movies reviews. Here we use "Ce commentaire est {}." as the hypothesis template and "positif" and "négatif" as candidate labels.
|
57 |
|
58 |
-
| **[allocine](https://huggingface.co/datasets/allocine)** | **time (ms)** | **MCC (x100)** |
|
59 |
-
| :--------------: | :-----------: | :------------: |
|
60 |
-
| [cmarkea/distilcamembert-base-nli](https://huggingface.co/cmarkea/distilcamembert-base-nli) | **205.54** | 63.71 |
|
61 |
-
| [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) | 378.39 | **73.74** |
|
62 |
-
| [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) | 520.58 | 70.05 |
|
63 |
|
64 |
The second one: [mlsum](https://huggingface.co/datasets/mlsum) used to train the summarization models. We use the articles summary part to predict their topics. In this aim, we aggregate sub-topics and select a few of them. In this case, the hypothesis template used is "C'est un article traitant de {}." and the candidate labels are: "économie", "politique", "sport", "cinéma", "musique" and "science".
|
65 |
|
66 |
-
| **[mlsum](https://huggingface.co/datasets/mlsum)** | **time (ms)** | **MCC (x100)** |
|
67 |
-
| :--------------: | :-----------: | :------------: |
|
68 |
-
| [cmarkea/distilcamembert-base-nli](https://huggingface.co/cmarkea/distilcamembert-base-nli) | **261.99** | 60.12 |
|
69 |
-
| [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) | 499.45 | **60.14** |
|
70 |
-
| [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) | 591.34 | 56.06 |
|
71 |
|
72 |
How to use DistilCamemBERT-Sentiment
|
73 |
------------------------------------
|
|
|
39 |
Benchmark
|
40 |
---------
|
41 |
|
42 |
+
We compare the [DistilCamemBERT](https://huggingface.co/cmarkea/distilcamembert-base) model with 2 other modelizations working on french language. The first one [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) is based on well named [CamemBERT](https://huggingface.co/camembert-base), the french RoBERTa model and the second one [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) based on [mDeBERTav3](https://huggingface.co/microsoft/mdeberta-v3-base) a multilingual model. To compare the performances the metrics of accuracy and [MCC (Matthews Correlation Coefficient)](https://en.wikipedia.org/wiki/Phi_coefficient) was used and for the mean inference time measure, an **AMD Ryzen 5 4500U @ 2.3GHz with 6 cores** was used:
|
43 |
|
44 |
+
| **NLI** | **time (ms)** | **accuracy (%)** | **MCC (x100)** |
|
45 |
+
| :--------------: | :-----------: | :--------------: | :------------: |
|
46 |
+
| [cmarkea/distilcamembert-base-nli](https://huggingface.co/cmarkea/distilcamembert-base-nli) | **51.35** | 77.45 | 66.24 |
|
47 |
+
| [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) | 105.0 | 81.72 | 72.67 |
|
48 |
+
| [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) | 299.18 | **83.43** | **75.15** |
|
49 |
|
50 |
Zero-shot classification
|
51 |
------------------------
|
|
|
55 |
|
56 |
For this part, we use 2 datasets, the first one: [allocine](https://huggingface.co/datasets/allocine) used to train the sentiment analysis models. The dataset is composed of 2 classes: "positif" and "négatif" appreciation of movies reviews. Here we use "Ce commentaire est {}." as the hypothesis template and "positif" and "négatif" as candidate labels.
|
57 |
|
58 |
+
| **[allocine](https://huggingface.co/datasets/allocine)** | **time (ms)** | **accuracy (%)** | **MCC (x100)** |
|
59 |
+
| :--------------: | :-----------: | :--------------: | :------------: |
|
60 |
+
| [cmarkea/distilcamembert-base-nli](https://huggingface.co/cmarkea/distilcamembert-base-nli) | **205.54** | 80.59 | 63.71 |
|
61 |
+
| [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) | 378.39 | **86.37** | **73.74** |
|
62 |
+
| [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) | 520.58 | 84.97 | 70.05 |
|
63 |
|
64 |
The second one: [mlsum](https://huggingface.co/datasets/mlsum) used to train the summarization models. We use the articles summary part to predict their topics. In this aim, we aggregate sub-topics and select a few of them. In this case, the hypothesis template used is "C'est un article traitant de {}." and the candidate labels are: "économie", "politique", "sport", "cinéma", "musique" and "science".
|
65 |
|
66 |
+
| **[mlsum](https://huggingface.co/datasets/mlsum)** | **time (ms)** | **accuracy (%)** | **MCC (x100)** |
|
67 |
+
| :--------------: | :-----------: | :--------------: | :------------: |
|
68 |
+
| [cmarkea/distilcamembert-base-nli](https://huggingface.co/cmarkea/distilcamembert-base-nli) | **261.99** | | 60.12 |
|
69 |
+
| [BaptisteDoyen/camembert-base-xnli](https://huggingface.co/BaptisteDoyen/camembert-base-xnli) | 499.45 | | **60.14** |
|
70 |
+
| [MoritzLaurer/mDeBERTa-v3-base-mnli-xnli](https://huggingface.co/MoritzLaurer/mDeBERTa-v3-base-mnli-xnli) | 591.34 | | 56.06 |
|
71 |
|
72 |
How to use DistilCamemBERT-Sentiment
|
73 |
------------------------------------
|