Muzhen Gaming
Muzhen Gaming
> I think this code can do it (untested): > > ``` > # Malus for too many active runs > # We scale based on total TP across all...
To summarise with a few examples: | Scenario | PR | Master | | --- | --- | --- | | 1x50% STC | 50*0.97 = 49 itp per test...
> @XInTheDark > You miss understand. for my formula we doesn't use the total itp (its not even needed to get the desired effect). Simply use number of running tests....
@locutus2 The specific problem observed with the simpler version is when there is a large number of tests, each with a low TP. For example 10x10% TP - master gives...
@locutus2 In the scenario that you mentioned, the 50% test will still have half the TP of the 100% test. The multiplier is based on total TP, so it’s the...
In fact in this specific example, my PR gives 64/32 itp respectively, for a total of 288 itp. but your version gives just 50/25, for a total of 225 itp....
The concept looks fine, but I think it's too complicated to have two parameters to calculate the TP. Also, I think the problem of not getting exactly half the cores...
thank you for the PR! will test it out with the openAI API and get back to you.
Thanks for the patch! I couldn't reproduce the issue since the JSON chunks seem to be nicely sent over here. Everything works normally for me anyway. Can you check if...
merged, thank you for your contribution!