should we bump this up to maybe 2K steps? e.g. mace-mp-0a finishes in ~800 for me so could be a little tight on some models finishing under the threshold?
The downside to this is for models that explode, they'll end up doing 2000 potentially very slow steps... I think merge this, but can I suggest we look at a (general utility) function we attach to dynamics that interrupts if forces become say 10e6?
Originally posted by @ElliottKasoar in #669 (comment)
Originally posted by @ElliottKasoar in #669 (comment)