ai-agents-fsdp-checkpoint-corr-using-fullyshardeddataparallel-summon-full-params--59ae08d9

THEN Remove or avoid the `summon_full_params` context during FSDP training if checkpoint integrity is required. Perform evaluation inference in a separate process after training, or switch to DDP (DistributedDataParallel) which does not exhibit this bug. If you must run inference during training, consider saving the model state before the call and restoring it afterward, though this may not be reliable.

# Problematic pattern inside a Trainer callback
with torch.no_grad():
    with fsdp.FullyShardedDataParallel.summon_full_params(model):
        outputs = model.generate(...)  # corrupts checkpoint

# Workaround: perform inference in a separate evaluation run,
# or use a deep copy of the model state if necessary.