Add TF weights (#1)

- Add TF weights (541601d8564e789a51a9e40df89831b87a609a8f)


Co-authored-by: Joao Gante <joaogante@users.noreply.huggingface.co>

README.md

@@ -10,16 +10,16 @@ tags:
 license: mit
 pipeline_tag: zero-shot-classification
 datasets: 
-- xlni
+- xnli
 metrics: 
 - accuracy
 ---
 
-# camembert-base-xlni
+# camembert-base-xnli
 
 ## Model description
 
-Camembert-base model fine-tuned on french part of XLNI dataset. <br>
+Camembert-base model fine-tuned on french part of XNLI dataset. <br>
 One of the few Zero-Shot classification model working on french 🇫🇷
 
 ## Intended uses & limitations
@@ -46,7 +46,7 @@ classifier(sequence, candidate_labels, hypothesis_template=hypothesis_template)
 ```
 
 - As a premise/hypothesis checker : <br>
-The idea is here to compute a probability of the form \\(P(premise|hypothesis)\\)
+The idea is here to compute a probability of the form \\( P(premise|hypothesis ) \\)
 
 ```python
 # load model and tokenizer
@@ -70,7 +70,7 @@ prob_label_is_true[0].tolist() * 100
 
 ## Training data
 
-Training data is the french fold of the [XLNI](https://research.fb.com/publications/xnli-evaluating-cross-lingual-sentence-representations/) dataset released in 2018 by Facebook. <br>
+Training data is the french fold of the [XNLI](https://research.fb.com/publications/xnli-evaluating-cross-lingual-sentence-representations/) dataset released in 2018 by Facebook. <br>
 Available with great ease using the ```datasets``` library :
 
 ```python
@@ -86,6 +86,7 @@ Main training parameters :
 - ```num_train_epochs = 4```
 - ```batch_size = 12``` (limited by GPU-memory)
 - ```weight_decay = 0.01```
+- ```metric_for_best_model = "eval_accuracy"```
 
 ## Eval results
 

config.json

@@ -11,16 +11,16 @@
   "hidden_dropout_prob": 0.1,
   "hidden_size": 768,
   "id2label": {
-    "0": "contradiction",
+    "0": "entailment",
     "1": "neutral",
-    "2": "entailment"
+    "2": "contradiction"
   },
   "initializer_range": 0.02,
   "intermediate_size": 3072,
   "label2id": {
-    "contradiction": 0,
+    "entailment": 0,
     "neutral": 1,
-    "entailment": 2
+    "contradiction": 2
   },
   "layer_norm_eps": 1e-05,
   "max_position_embeddings": 514,
@@ -30,7 +30,7 @@
   "output_past": true,
   "pad_token_id": 1,
   "position_embedding_type": "absolute",
-  "transformers_version": "4.4.2",
+  "transformers_version": "4.3.3",
   "type_vocab_size": 1,
   "use_cache": true,
   "vocab_size": 32005