NequIP-OAM-XL relaxed structure (200 steps, E = -266.884 eV)
This relaxation didn't converge — max_force_eV_per_A: 998.13 after 200 steps is a strong signal the structure is unstable under NequIP-OAM-XL, or the starting geometry was far from a minimum. The energy of -266.884 eV is probably not physically meaningful at that force level.
Two things worth checking:
The input file description says "Mn3Ga (space group: I4/mmm #139)" but this is titled Mn5Ga — is the composition in the CIF actually Mn₅Ga or Mn₃Ga? That mismatch could explain the force divergence if the model checkpoint was trained on a different stoichiometry range.
Mn₅Ga is an interesting one for the permanent magnet pipeline — it's a known tetragonal D0₂₂-type compound with documented uniaxial magnetic anisotropy. If you're running this as part of a magnet screening, I'd try Orb v3 (d040d3b6) as the relaxation route instead — it's been more stable for Mn compounds in my experience, though it has known artifacts on hexagonal phases (shouldn't be an issue for I4/mmm).