push farasa base model

.gitattributes

@@ -14,3 +14,5 @@
 *.pb filter=lfs diff=lfs merge=lfs -text
 *.pt filter=lfs diff=lfs merge=lfs -text
 *.pth filter=lfs diff=lfs merge=lfs -text
+pytorch_model.bin filter=lfs diff=lfs merge=lfs -text
+model.ckpt.data-00000-of-00001 filter=lfs diff=lfs merge=lfs -text

README.md

@@ -0,0 +1,84 @@
+---
+language: ar
+tags:
+- pytorch
+- tf
+- QARiB
+- qarib
+datasets:
+- arabic_billion_words
+- open_subtitles
+- twitter
+- Farasa
+metrics:
+- f1
+widget:
+ - text: "و+قام ال+مدير [MASK]"
+---
+# QARiB: QCRI Arabic and Dialectal BERT
+## About QARiB Farasa
+QCRI Arabic and Dialectal BERT  (QARiB) model, was trained on a collection of ~ 420 Million tweets and ~ 180 Million sentences of text.
+For the tweets, the data was collected using twitter API and using language filter. `lang:ar`. For the text data, it was a combination from
+[Arabic GigaWord](url), [Abulkhair Arabic Corpus]() and [OPUS](http://opus.nlpl.eu/).
+QARiB: Is the Arabic name for "Boat".
+## Model and Parameters:
+- Data size: 14B tokens 
+- Vocabulary: 64k
+- Iterations:  10M
+- Number of Layers: 12
+## Training QARiB
+See details in [Training QARiB](https://github.com/qcri/QARIB/Training_QARiB.md)
+## Using QARiB
+You can use the raw model for either masked language modeling or next sentence prediction, but it's mostly intended to be fine-tuned on a downstream task. See the model hub to look for fine-tuned versions on a task that interests you. For more details, see [Using QARiB](https://github.com/qcri/QARIB/Using_QARiB.md)
+
+This model expects the data to be segmented. You may use [Farasa Segmenter](https://farasa-api.qcri.org/segmentation/) API.
+
+### How to use
+You can use this model directly with a pipeline for masked language modeling:
+```python
+>>>from transformers import pipeline
+>>>fill_mask = pipeline("fill-mask", model="./models/bert-base-qarib_far")
+>>> fill_mask("و+قام ال+مدير [MASK]")
+[
+{'sequence': '[CLS] وقام المدير بالعمل [SEP]', 'score': 0.0678194984793663, 'token': 4230, 'token_str': 'بالعمل'}, 
+{'sequence': '[CLS] وقام المدير بذلك [SEP]', 'score': 0.05191086605191231, 'token': 984, 'token_str': 'بذلك'}, 
+{'sequence': '[CLS] وقام المدير بالاتصال [SEP]', 'score': 0.045264165848493576, 'token': 26096, 'token_str': 'بالاتصال'}, 
+{'sequence': '[CLS] وقام المدير بعمله [SEP]', 'score': 0.03732728958129883, 'token': 40486, 'token_str': 'بعمله'}, 
+{'sequence': '[CLS] وقام المدير بالامر [SEP]', 'score': 0.0246378555893898, 'token': 29124, 'token_str': 'بالامر'}
+]
+>>> fill_mask("و+قام+ت ال+مدير+ة [MASK]")
+[{'sequence': '[CLS] وقامت المديرة بذلك [SEP]', 'score': 0.23992691934108734, 'token': 984, 'token_str': 'بذلك'}, 
+{'sequence': '[CLS] وقامت المديرة بالامر [SEP]', 'score': 0.108805812895298, 'token': 29124, 'token_str': 'بالامر'}, 
+{'sequence': '[CLS] وقامت المديرة بالعمل [SEP]', 'score': 0.06639821827411652, 'token': 4230, 'token_str': 'بالعمل'}, 
+{'sequence': '[CLS] وقامت المديرة بالاتصال [SEP]', 'score': 0.05613093823194504, 'token': 26096, 'token_str': 'بالاتصال'}, 
+{'sequence': '[CLS] وقامت المديرة المديرة [SEP]', 'score': 0.021778125315904617, 'token': 41635, 'token_str': 'المديرة'}]
+>>> fill_mask("قللي وشفيييك يرحم [MASK]")
+[{'sequence': '[CLS] قللي وشفيييك يرحم والديك [SEP]', 'score': 0.4152909517288208, 'token': 9650, 'token_str': 'والديك'}, 
+{'sequence': '[CLS] قللي وشفيييك يرحملي [SEP]', 'score': 0.07663793861865997, 'token': 294, 'token_str': '##لي'}, 
+{'sequence': '[CLS] قللي وشفيييك يرحم حالك [SEP]', 'score': 0.0453166700899601, 'token': 2663, 'token_str': 'حالك'}, 
+{'sequence': '[CLS] قللي وشفيييك يرحم امك [SEP]', 'score': 0.04390475153923035, 'token': 1942, 'token_str': 'امك'}, 
+{'sequence': '[CLS] قللي وشفيييك يرحمونك [SEP]', 'score': 0.027349254116415977, 'token': 3283, 'token_str': '##ونك'}]
+```
+## Evaluations:
+|**Experiment** |**mBERT**|**AraBERT0.1**|**AraBERT1.0**|**ArabicBERT**|**QARiB**|
+|---------------|---------|--------------|--------------|--------------|---------|
+|Dialect Identification | 6.06% | 59.92% | 59.85% | 61.70% | **65.21%** |
+|Emotion Detection | 27.90% | 43.89% | 42.37% | 41.65% | **44.35%** |
+|Named-Entity Recognition (NER) | 49.38% | 64.97% | **66.63%** | 64.04% | 61.62% |
+|Offensive Language Detection | 83.14% | 88.07% | 88.97% | 88.19% | **91.94%** |
+|Sentiment Analysis | 86.61% | 90.80% | **93.58%** | 83.27% | 93.31% |
+## Model Weights and Vocab Download
+From Huggingface site: https://huggingface.co/qarib/bert-base-qarib_far
+## Contacts
+Ahmed Abdelali, Sabit Hassan, Hamdy Mubarak, Kareem Darwish and Younes Samih
+## Reference
+```
+@article{abdelali2021pretraining,
+    title={Pre-Training BERT on Arabic Tweets: Practical Considerations},
+    author={Ahmed Abdelali and Sabit Hassan and Hamdy Mubarak and Kareem Darwish and Younes Samih},
+    year={2021},
+    eprint={2102.10684},
+    archivePrefix={arXiv},
+    primaryClass={cs.CL}
+}
+```

config.json

@@ -0,0 +1,19 @@
+{
+  "attention_probs_dropout_prob": 0.1,
+  "directionality": "bidi",
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "max_position_embeddings": 512,
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pooler_fc_size": 768,
+  "pooler_num_attention_heads": 12,
+  "pooler_num_fc_layers": 3,
+  "pooler_size_per_head": 128,
+  "pooler_type": "first_token_transform",
+  "type_vocab_size": 2,
+  "vocab_size": 64000
+}

model.ckpt.data-00000-of-00001

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e78e8c1df6e1ad6ff226c4f1a2366264fbb53479fabc3783a8a9dd8663f226e4
+size 1630212128

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:32c0d5c9228748942e17413080b9eeca1d7869de6c8d8e9b96daf76216649e5d
+size 543488365