blip support for training (#21021)

* `blip` support for training

* remove labels creation

* remove unneeded `decoder_input_ids` creation

* final changes

- add colab link to documentation
- reduction = mean for loss

* fix nits

* update link

* clearer error message

docs/source/en/model_doc/blip.mdx

@@ -31,6 +31,10 @@ However, most existing pre-trained models only excel in either understanding-bas
 This model was contributed by [ybelkada](https://huggingface.co/ybelkada).
 The original code can be found [here](https://github.com/salesforce/BLIP).
 
+## Resources
+
+- [Jupyter notebook](https://github.com/huggingface/notebooks/blob/main/examples/image_captioning_blip.ipynb) on how to fine-tune BLIP for image captioning on a custom dataset
+
 
 ## BlipConfig