add doc of converting t5 to hg model (#96)

docs/convert_t5_to_hf.md

@@ -0,0 +1,24 @@
+This document describes how to convert our t5 to huggingface model and then upload to remote
+### Load T5 Tensorflow model to transformers.
+```python
+from transformers import T5Config, T5Tokenizer, T5ForConditionalGeneration
+# load t5 tf model to transformers model
+config = T5Config.from_pretrained('t5-base')
+model = T5ForConditionalGeneration.from_pretrained("path/to/model.ckpt-1011000.index", from_tf=True, config=config)
+
+# get t5 tokenizer ready too
+tokenizer = T5Tokenizer.from_pretrained('t5-base')
+
+# save model & tokenizer together to the folder to upload to huggingface
+model.save_pretrained("path/to/model/")
+tokenizer.save_pretrained("path/to/model/")
+```
+
+### Using `transformers-cli` to upload the model folder to huggingface
+1. Use `transformers-cli login` to login
+2. Upload the model to huggingface under our organization
+```
+transformers-cli upload path/to/model/ --organization castorini
+```
+
+Please see [transformers](https://huggingface.co/transformers/model_sharing.html) for details.