Checkpoint is manipulated

[elwlwlwk]

https://wandb.ai/macrocosmos/folding-openmm/runs/ig30rwe2/overview

/vali-s3-demo-do/outputs/2qdv/5HYk8DMK/5GehSP5B/2025-01-19_11-49-33/md_0_1.cpt
this checkpoint file has potential energy around -5200. but load that checkpoint and save it's state using simulation.saveState and re-run from state file, potential energy is around -1200

pdb = app.PDBFile(f"{pdb_id}/{pdb_file}.pdb")
simulation, _ = OpenMMSimulation().create_simulation(
    pdb=pdb,
    system_config=config
)

simulation.loadCheckpoint(f"{pdb_id}/md_0_1.cpt")
simulation.saveState(f"{pdb_id}/state_{sim_id}.xml")

checkpoint_energies = []
for i in range(100):
    simulation.step(100)
    energy = simulation.context.getState(getEnergy=True).getPotentialEnergy()._value
    checkpoint_energies.append(energy)
print(f"Checkpoint Energies: {checkpoint_energies}")

simulation, _ = OpenMMSimulation().create_simulation(
    pdb=pdb,
    system_config=config
)

simulation.loadState(f"{pdb_id}/state_{sim_id}.xml")
state_energies = []
for i in range(100):
    simulation.step(100)
    energy = simulation.context.getState(getEnergy=True).getPotentialEnergy()._value
    state_energies.append(energy)
print(f"State Energies: {state_energies}")

[mccrindlebrian]

We are continuing to work on this issue as we are finding that the energies are not always increasing on the state-level. Therefore, they will fail the gradient check. As such, we are going to add a different