Feature Request: Way to kill an inference without killing the server

[paoletto]

Prerequisites

• I am running the latest code. Mention the version if possible as well.
• I carefully followed the README.md.
• I searched using keywords relevant to my issue to make sure that I am creating a new issue that is not already open (or closed).
• I reviewed the Discussions, and have a new and useful enhancement to share.

Feature Description

Curl'ed inference requests occasionally get stuck, either in llama-server or in whisper-server.
as other requests are queued, killing the server would make every other queued requests fail.
So i wonder: how to kill one of the ongoing tasks (could be multiple if the server is set to use more than 1 concurrent job), without having to actually kill the server, and so letting the queued task move on?

Motivation

Requests occasionally get stuck and block the server

Possible Implementation

additional API endpoint taking task id?

[paoletto]

could be duplicate of #6421, although a dedicated way to kill current jobs i believe could have merit on its own

[ngxson]

I mentioned the same problem a while ago here: #9273

Will revisit it in a few days.