Iterations number for a model

[Atulsingh92]

I have a model (in emukit, basic multi-fidelity application), where I have the following after several m.optimize()

Optimization restart 493/500, f = 561.0128272525858
Optimization restart 494/500, f = 463.1343825925499
Optimization restart 495/500, f = 448.1619253231641
Optimization restart 496/500, f = 362.06335957792425
Optimization restart 497/500, f = 195.34076107170472
Optimization restart 498/500, f = 455.30806006647487
Warning - optimization restart 499/500 failed
Optimization restart 500/500, f = 366.488550073236

and in another similar case, where I have fewer restarts, it is as follows,

Optimization restart 6/10, f = 460.77444487689746
Optimization restart 7/10, f = 338.2667043881471
 C:\Users\aus1n19\anaconda3\envs\localenv\lib\site-packages\GPy\kern\src\stationary.py:146: RuntimeWarning:overflow encountered in multiply
 C:\Users\aus1n19\anaconda3\envs\localenv\lib\site-packages\GPy\kern\src\stationary.py:146: RuntimeWarning:overflow encountered in add
 C:\Users\aus1n19\anaconda3\envs\localenv\lib\site-packages\GPy\kern\src\stationary.py:584: RuntimeWarning:invalid value encountered in subtract
Optimization restart 8/10, f = 369.0709437934151
Optimization restart 9/10, f = 462.47411111313573
Optimization restart 10/10, f = 369.66244558526535

My question is, does this value should always be negative (as in the examples), and should be converging (also as in examples)?
Is there anything that I could do to improve this?

(PS: My multi-fidelity model, has 79 data points for Lowfi and 4 points for high-fi, so most likely there is not enough data for the difference model, but) Could the above m.optimze step be corrected elsewise, so that I have a converging value in it, assuming that's what one should aim for?

[Atulsingh92]

Corrected this, by normalizing the output, as well.