Update README.md

README.md

@@ -16,6 +16,15 @@ It was trained on both T5 (text) and the [AnimaTextToImagePipeline](https://hugg
 
 *The project is a work in progress, see the image above.*
 
+## Inference
+
+```python
+bert = CosmosBert.from_pretrained('nightknocker/cosmos-bert')
+tokenizer = AutoTokenizer.from_pretrained('nightknocker/cosmos-bert')
+inputs = tokenizer(text, return_tensors='pt').to('cuda')
+crossattn_emb = bert.forward(**inputs, return_dict=True).last_hidden_state
+```
+
 ## References
 
 - [Recurrent Qwen](https://huggingface.co/nightknocker/recurrent-qwen3-z-image-turbo)