---
title: "Discussion on \"Bound the Work Before the Model Runs: Gating On-Device Inference in TikTok's Android Client\" | Hashnode"
description: "Discussion on \"Bound the Work Before the Model Runs: Gating On-Device Inference in TikTok's Android Client\". An 8% visibility gate, a 100 ms dwell timer, bounded vision input and an AICore-to-ByteNN fallback in TikTok's Android client."
url: "https://hashnode.com/posts/bound-the-work-before-the-model-runs-gating-on-device-inference-in-tiktok-s-android-client/6afe99b244b5a86d91fd652e"
---

[Feed](https://hashnode.com/feed) /

Discussion

# [Bound the Work Before the Model Runs: Gating On-Device Inference in TikTok's Android Client](https://sairamkotha.hashnode.dev/bound-the-work-before-the-model-runs-gating-on-device-inference-in-tiktok-s-android-client?utm_source=hashnode&utm_medium=feed)

An 8% visibility gate, a 100 ms dwell timer, bounded vision input and an AICore-to-ByteNN fallback in TikTok's Android client.

15h ago · 10 min read · sairamkotha.hashnode.dev  
Read the article

-   [#android](https://hashnode.com/tag/android)
-   [#reverse-engineering](https://hashnode.com/tag/reverse-engineering)
-   [#mobile-development](https://hashnode.com/tag/mobile-development)
-   [#machine-learning](https://hashnode.com/tag/machine-learning)
-   [#performance](https://hashnode.com/tag/performance)

## Written by

[Sai Ram Kotha](https://hashnode.com/@sairamkotha)

@sairamkotha · Bengaluru, India

  
[View profile](https://hashnode.com/@sairamkotha)

AI, Mobile

## Responses

Markdown supported · ⌘↵ to post

No responses yet.
