Enable stable-fast pretracing in T2I, I2I and upscale pipelines [30 LPT]

[rickstaa]

Overview

Since its implementation, the stable-fast optimization has enhanced the image-to-video pipeline by providing a 50% speed-up. The primary drawback is that it utilizes dynamic pre-tracing to optimize requests, causing a longer response time on the first try. However, if the first request involves the largest possible image size, all smaller image sizes will also be pre-traced.

This optimization feature was initially not added to the text-to-image, image-to-image, and upscale pipelines due to the support for batch requests at that time, which required pre-tracing each batch size. Since then, we've transitioned from batch requests to sequential requests to mitigate memory issues and improve performance. Consequently, pre-tracing for these pipelines has become feasible again 🚀.

We are seeking a dedicated community member to apply the existing code from this pull request to the text-to-image, image-to-image, and upscale pipelines. Completing this bounty will significantly enhance network performance by enabling orchestrators to achieve a 50% speed-up in these pipelines ⚡🙏.

Required Skillset

• Familiarity with FastAPI.
• Python programming skills.

Bounty Requirements

Bounty Requirements

1. Pre-Tracing Implementation:
   • For each pipeline (text-to-image, image-to-image, and upscale), ensure that when orchestrators have the SFAST setting enabled, the largest image size request is pre-traced during startup.
   • Implement this by setting SFAST_WARMUP=true, ensuring orchestrators achieve the quickest response times.
2. Performance Comparison Table:
   • Create a comparison table for each pipeline.
   • The table should include:
     • Pre-trace time
     • Resulting inference time
     • Comparison with non-optimized time

Implementation Tips

• Utilize Earlier Work: Don't forget to review the code already provided by @rickstaa in an earlier pull request and the pre-tracing code already present in the image-to-video pipeline. This can provide valuable insights and a foundation for your work.
• Utilize Developer Documentation: Check out our developer documentation for the worker and runner.

How to Apply

1. Express Your Interest: Comment on this issue to indicate your interest and explain why you're the ideal candidate for the task.
2. Wait for Review: Our team will review expressions of interest and select the best candidate.
3. Get Assigned: If selected, we'll assign the GitHub issue to you.
4. Start Working: Dive into your task! If you need assistance or guidance, comment on the issue or join the discussions in the #🛋│developer-lounge channel on our Discord server.
5. Submit Your Work: Create a pull request in the relevant repository and request a review.
6. Notify Us: Comment on this GitHub issue when your pull request is ready for review.
7. Receive Your Bounty: We'll arrange the bounty payment once your pull request is approved.
8. Gain Recognition: Your valuable contributions will be showcased in our project's changelog.

Thank you for your interest in contributing to our project 💛!

[!WARNING] Please wait for the issue to be assigned to you before starting work. To prevent duplication of effort, submissions for unassigned issues will not be accepted.

[rickstaa]

This bounty was implemented by @lukiod in https://github.com/livepeer/ai-worker/pull/134.

[JJassonn69]

This bounty was implemented by @lukiod in livepeer/ai-worker#134.

I think you mean @JJassonn69.