# Converting Python api from rasa nlu 0.9 to rasa 3.\* **URL:** \ **Category:** Rasa Open Source\ **Created:** [April 3, 2022, 2:54am UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336 "2022-04-03T02:54:35Z")\ **Posts on this page:** 8\ **Page:** 1
**Author:** ![ravisripada\_2022](https://avatars.discourse-cdn.com/v4/letter/r/46a35a/32.png) [@ravisripada\_2022](https://forum.rasa.com/u/ravisripada_2022)\ **Post date:** [April 3, 2022, 2:54am UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/1 "2022-04-03T02:54:35Z")
Currently we have a feature in our web application that user rasa nlu from rasa nlu 0.9. Rasa NLU is used for search rather than chat. I was trying to see if there are similar libraries in rasa 3.\* Since rasa 1.\* it appears that we should have converted to using http api or nlu only server. What is the best way to upgrade our current features to rasa 3.\* version. Is there a way to still use python api? ---
**Author:** ![amn41](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/amn41/32/20_2.png) [@amn41](https://forum.rasa.com/u/amn41)\ **Post date:** [April 4, 2022, 7:19am UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/2 "2022-04-04T07:19:01Z")
Hi Ravi - yes you can still use the python API to build apps. Possibly a good way to discover the relevant high level functions is to look at how the [cli](https://rasa.com/docs/rasa/command-line-interface) commands are [implemented](https://github.com/RasaHQ/rasa/tree/main/rasa/cli) ---
**Author:** ![ravisripada\_2022](https://avatars.discourse-cdn.com/v4/letter/r/46a35a/32.png) [@ravisripada\_2022](https://forum.rasa.com/u/ravisripada_2022)\ **Post date:** [April 6, 2022, 10:16pm UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/3 "2022-04-06T22:16:07Z")
Thanks Alan. I have one more question: We were json format for generating our training data. I don’t see a mention of json formats in documentation since 2.\*. Is Json for training data still supported? or do we need to create a yaml file? ---
**Author:** ![amn41](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/amn41/32/20_2.png) [@amn41](https://forum.rasa.com/u/amn41)\ **Post date:** [April 7, 2022, 8:01am UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/4 "2022-04-07T08:01:59Z")
yes, we’ve deprecated the json format in favour of yaml. ---
**Author:** ![ravisripada\_2022](https://avatars.discourse-cdn.com/v4/letter/r/46a35a/32.png) [@ravisripada\_2022](https://forum.rasa.com/u/ravisripada_2022)\ **Post date:** [April 12, 2022, 4:26pm UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/5 "2022-04-12T16:26:55Z")
Thanks Alan. I was able to upgrade and app is working as intended. One more question: I see that model file is like over 100GB in size is there a way to reduce this? ---
**Author:** ![amn41](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/amn41/32/20_2.png) [@amn41](https://forum.rasa.com/u/amn41)\ **Post date:** [April 13, 2022, 11:32am UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/6 "2022-04-13T11:32:17Z")
wow, that is huge! can you try unzipping the model folder and seeing which files are so big? ---
**Author:** ![ravisripada\_2022](https://avatars.discourse-cdn.com/v4/letter/r/46a35a/32.png) [@ravisripada\_2022](https://forum.rasa.com/u/ravisripada_2022)\ **Post date:** [April 14, 2022, 2:05am UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/7 "2022-04-14T02:05:31Z")
My bad the size was 170 MB. The file that is big: DIETClassifier.tf\_model.data-00000-of-00001 ---
**Author:** ![amn41](https://dub1.discourse-cdn.com/flex013/user_avatar/forum.rasa.com/amn41/32/20_2.png) [@amn41](https://forum.rasa.com/u/amn41)\ **Post date:** [April 14, 2022, 7:21am UTC](https://forum.rasa.com/t/converting-python-api-from-rasa-nlu-0-9-to-rasa-3/52336/8 "2022-04-14T07:21:02Z")
ok yeah that’s more reasonable. If you want to get a smaller model, you can play around with the DIET hyperparameters. We have a bunch of great resources on youtube like the [algorithm whiteboard](https://rasa.com/blog/nlp-for-developers-and-the-algorithm-whiteboard/) , which can help you understand how the model works and make informed decisions