paddler icon indicating copy to clipboard operation
paddler copied to clipboard

(stastsd) not wokring idle slot with latest llama.cpp

Open bodybreaker opened this issue 1 year ago • 4 comments

paddler_slots_idle not working with lastest llama.cpp

bodybreaker avatar Aug 28 '24 08:08 bodybreaker

The slot states were moved from the /health endpoint to a new /slots endpoint. See https://github.com/ggerganov/llama.cpp/pull/9056.

us58 avatar Sep 01 '24 22:09 us58

@us58 Thanks for letting me know. I am working on the update now.

I also mentioned the issue in llama.cpp discussions, such breaking change was kind of unexpected: https://github.com/ggerganov/llama.cpp/discussions/9276

mcharytoniuk avatar Sep 02 '24 12:09 mcharytoniuk

@bodybreaker @us58 I released a new version that uses the new /slots endpoint from llama.cpp server: https://github.com/distantmagic/paddler/releases/tag/v0.7.0

Please let me know if that solved StatsD issue.

mcharytoniuk avatar Sep 04 '24 01:09 mcharytoniuk

I think I might be having a problem related to this. In my llama.cpp logs I see the monitoring requests coming in from the paddler agent, but they're still coming into the /health endpoint instead of the /slots endpoint as they should be. This is with release 0.8.0 so as I understand it this should be working?

debrisapron avatar Sep 26 '24 22:09 debrisapron

@debrisapron @bodybreaker I have just released a new stable version of Paddler (v1.0.0) and changed the CLI framework, overall it underwent a total rewrite.

I think your issue should be solved now, if it still persists feel free to reopen (please check the README though, some flag names have changed).

mcharytoniuk avatar Nov 20 '24 20:11 mcharytoniuk