Update README.md

README.md

@@ -1,3 +1,50 @@
 ---
 license: mit
 ---
+
+This model is a RoBERTa model trained on a programming language code - WolfSSL.
+
+The programming language is C/C++, but the actual inference can also use other languages. 
+
+Using the model to unmask can be done in the following way
+
+```python
+from transformers import pipeline
+unmasker = pipeline('fill-mask', model='mstaron/wolfBERTa')
+unmasker("Hello I'm a <mask> model.")
+```
+
+To obtain the embeddings for downstream task can be done in the following way:
+
+```python
+# import the model via the huggingface library
+from transformers import AutoTokenizer, AutoModelForMaskedLM
+
+# load the tokenizer and the model for the pretrained wolfBERTa
+tokenizer = AutoTokenizer.from_pretrained('mstaron/wolfBERTa')
+
+# load the model
+model = AutoModelForMaskedLM.from_pretrained("mstaron/wolfBERTa")
+
+# import the feature extraction pipeline
+from transformers import pipeline
+
+# create the pipeline, which will extract the embedding vectors
+# the models are already pre-defined, so we do not need to train anything here
+features = pipeline(
+    "feature-extraction",
+    model=model,
+    tokenizer=tokenizer, 
+    return_tensor = False
+)
+
+# extract the features == embeddings
+lstFeatures = features('Class HTTP::X1')
+
+# print the first token's embedding [CLS]
+# which is also a good approximation of the whole sentence embedding
+# the same as using np.mean(lstFeatures[0], axis=0)
+lstFeatures[0][0]
+```
+
+In order to use the model, we need to train it on the downstream task.