init

README.md

@@ -0,0 +1,38 @@
+# PolitBERT
+
+## Background
+
+This model was created to specialize on political speeches, interviews and press briefings of English-speaking politicians.
+
+## Training
+The model was initialized using the pre-trained weights of BERT<sub>BASE</sub> and trained for 20 epochs on the standard MLM task with default parameters.
+The used learning rate was 5e-5 with a linearly decreasing schedule and AdamW.
+The used batch size is 8 per GPU while beeing trained on two Nvidia GTX TITAN X.
+The rest of the used configuration is the same as in ```AutoConfig.from_pretrained('bert-base-uncased')```.
+As a tokenizer the default tokenizer of BERT was used (```BertTokenizer.from_pretrained('bert-base-uncased')```)
+
+## Dataset
+PolitBERT was trained on the following dataset, which has been split up into single sentences:
+<https://www.kaggle.com/mauricerupp/englishspeaking-politicians>
+
+## Usage
+To predict a missing word of a sentence, the following pipeline can be applied:
+
+```
+from transformers import pipeline, BertTokenizer, AutoModel
+
+fill_mask = pipeline("fill-mask", 
+                     model=AutoModel.from_pretrained('maurice/PolitBERT'), 
+                     tokenizer=BertTokenizer.from_pretrained('bert-base-uncased'))
+
+print(fill_mask('Donald Trump is a [MASK].'))
+```
+
+## Training Results
+Evaluation Loss:
+![evalloss](evalloss_BERT.png)
+Training Loss:
+![evalloss](loss_BERT.png)
+Learning Rate Schedule:
+![evalloss](LR_BERT.png)
+

config.json

@@ -0,0 +1,21 @@
+{
+  "_name_or_path": "bert-base-uncased",
+  "architectures": [
+    "BertForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-12,
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 0,
+  "type_vocab_size": 2,
+  "vocab_size": 30522
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd89d77f92b6cc7ea4eeae1fbfa86d78d5477e871919fec352c3582837c3b813
+size 438147282

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ac5123eac1bfaa1ee8670a68373af816ddceca90e6bcf4102f6d1cb72df4ec5a
+size 1839