Update README.md

README.md

@@ -10,6 +10,8 @@ _Work in progress._
 
 Language model for inclusive language in German, fine-tuned on [mT5](https://arxiv.org/abs/2010.11934).
 
+An experimental model version is released [on Huggingface](https://huggingface.co/diversifix/diversiformer).
+
 ## Tasks
 
 - **DETECT**: Recognizes instances of the generic masculine, and of other exclusive language. To do.
@@ -24,9 +26,29 @@ Language model for inclusive language in German, fine-tuned on [mT5](https://arx
 
     ◀️ `Das wartende Kollegium wunderte sich.`
 
-## Dev ideas
+## Usage
+
+```python
+from transformers import T5Tokenizer, TFT5ForConditionalGeneration
+
+tokenizer = T5Tokenizer.from_pretrained("google/mt5-small")
+model = TFT5ForConditionalGeneration.from_pretrained("diversifix/diversiformer")
+
+def generate(prompt, tokenizer, model):
+    tokenized_text = tokenizer.encode(prompt, return_tensors="tf")
+    ids = model.generate(tokenized_text, max_length=500)
+    output = tokenizer.decode(ids[0], skip_special_tokens=True)
+    return output
+
+prompts = [
+    'Ersetze "Schüler" durch "Schülerin oder Schüler": Die Schüler kamen zu spät.',
+    'Ersetze "Lehrer" durch "Kollegium": Die wartenden Lehrer wunderten sich.',
+]
 
-- Use classifier to filter out training data of low quality? (~adversarial approach)
+for prompt in prompts:
+    output = generate(prompt, tokenizer, model)
+    print(f"{prompt}\n{output}\n\n")
+```
 
 ## License