# Creating Tflite model with user input

**URL:** <https://discuss.ai.google.dev/t/creating-tflite-model-with-user-input/25977>\
**Category:** General Discussion\
**Tags:** tflite\_micro, help\_request\
**Created:** [October 19, 2021, 1:38pm UTC](https://discuss.ai.google.dev/t/creating-tflite-model-with-user-input/25977 "2021-10-19T13:38:45Z")\
**Posts on this page:** 2\
**Page:** 1

<div class="post-metadata">

**Author:** ![Aditi\_Bhalawat](https://avatars.discourse-cdn.com/v4/letter/a/4491bb/32.png) [@Aditi\_Bhalawat](https://discuss.ai.google.dev/u/Aditi_Bhalawat)\
**Post date:** [October 19, 2021, 1:38pm UTC](https://discuss.ai.google.dev/t/creating-tflite-model-with-user-input/25977/1 "2021-10-19T13:38:45Z")

</div>

I want to create a tensor-flow Lite model in which the microcontroller(i.e. ESP32) initially learns from user input and latter when same condition or related condition carries it gives the output as provided by user.

Could you guide me how to perform that?

---

<div class="post-metadata">

**Author:** ![Fergus\_Henderson](https://d1dlmcr85iqnpo.cloudfront.net/user_avatar/discuss.ai.google.dev/fergus_henderson/32/10785_2.png) [@Fergus\_Henderson](https://discuss.ai.google.dev/u/Fergus_Henderson)\
**Post date:** [February 22, 2022, 2:47pm UTC](https://discuss.ai.google.dev/t/creating-tflite-model-with-user-input/25977/3 "2022-02-22T14:47:48Z")

</div>

Microcontrollers have limited resources, which makes things more difficult. There has been some academic work on online learning on microcontrollers, e.g. [TinyOL](https://arxiv.org/pdf/2103.08295.pdf). But I’m not aware of any open-source ML frameworks for doing ML training on microcontrollers.

In particular, for “TensorFlow Lite for MicroControllers”, according to the [docs](https://www.tensorflow.org/lite/microcontrollers#limitations), “On device training is not supported.”

So you might want to consider an architecture where the training happens on a server and the microcontroller occasionally fetches a new model from the server?