I tried to use K2.5 for 3 of the 6-ish days the model was available and didn't get anything out of it that I could use. There were just too many API errors. I guess the times of day when I was trying to use it was when everyone else was too.
To be precise, it wasn't a full week but about six days.
And yeah the context/token issue is something I noticed quickly, besides from various API errors where most commen "fetch failed" and "Uknown API error" appeared.
It struggled with a tbf quite complex 20k LOC project – where however there were no subfolders – where Codex did not complain at all and various times it produced – very obvious – crap on simple things. But it solved at least one issue.
It also heavily struggled with a 25k LOC Android project – where half was Java and other half XML – in which the creator sadly added multiple features per commit and I basically only wanted to have a reduced version of the app. So I told Kimi to use git and remove a feature that got just added. For that I already started a rebase on the last commit and theorethically it only would have needed to look at a few 100 LoC. In the first attempts it just threw API errors cause appearently it looked at the whole code base or something. Then I told it to just look at one file at a time and it did that but it took a bit too long imo.
I used Kimi 2.5 on VS Code IDE, which gives a great experience. Thank you!
I tried to use K2.5 for 3 of the 6-ish days the model was available and didn't get anything out of it that I could use. There were just too many API errors. I guess the times of day when I was trying to use it was when everyone else was too.
To be precise, it wasn't a full week but about six days.
And yeah the context/token issue is something I noticed quickly, besides from various API errors where most commen "fetch failed" and "Uknown API error" appeared.
It struggled with a tbf quite complex 20k LOC project – where however there were no subfolders – where Codex did not complain at all and various times it produced – very obvious – crap on simple things. But it solved at least one issue.
It also heavily struggled with a 25k LOC Android project – where half was Java and other half XML – in which the creator sadly added multiple features per commit and I basically only wanted to have a reduced version of the app. So I told Kimi to use git and remove a feature that got just added. For that I already started a rebase on the last commit and theorethically it only would have needed to look at a few 100 LoC. In the first attempts it just threw API errors cause appearently it looked at the whole code base or something. Then I told it to just look at one file at a time and it did that but it took a bit too long imo.