Lightning-AI/pytorch-lightning · error · MisconfigurationException

A single `Optimizer` cannot have multiple parameter groups w

Error message

A single `Optimizer` cannot have multiple parameter groups with identical `name` values. {name} has duplicated parameter group names {duplicates}

What it means

LearningRateMonitor derives names for each optimizer's param groups from an optional 'name' entry in the group dicts. If a single optimizer has multiple param groups whose `name` values collide, names would be ambiguous, so `_check_duplicates_and_update_name` raises MisconfigurationException listing the duplicates.

Source

Thrown at src/lightning/pytorch/callbacks/lr_monitor.py:364

    def _check_duplicates_and_update_name(
        self,
        optimizer: Optimizer,
        name: str,
        seen_optimizers: list[Optimizer],
        seen_optimizer_types: defaultdict[type[Optimizer], int],
        lr_scheduler_config: Optional[LRSchedulerConfig],
    ) -> list[str]:
        seen_optimizers.append(optimizer)
        optimizer_cls = type(optimizer)
        if lr_scheduler_config is None or lr_scheduler_config.name is None:
            seen_optimizer_types[optimizer_cls] += 1

        # Multiple param groups for the same optimizer
        param_groups = optimizer.param_groups
        duplicates = self._duplicate_param_group_names(param_groups)
        if duplicates:
            raise MisconfigurationException(
                "A single `Optimizer` cannot have multiple parameter groups with identical "
                f"`name` values. {name} has duplicated parameter group names {duplicates}"
            )

        name = self._add_prefix(name, optimizer_cls, seen_optimizer_types)
        names = [self._add_suffix(name, param_groups, i) for i in range(len(param_groups))]
        if self.log_key_prefix:
            names = [f"{self.log_key_prefix}{n}" for n in names]
        return names

View on GitHub (pinned to 9fed5c27d2)

Solutions

1. Give each param group a unique 'name' value (or remove the 'name' keys so Lightning auto-generates suffixed names)
2. If groups genuinely should share a name, merge them into one group
3. Check the `duplicates` list in the message against your `configure_optimizers` output

Example fix

# before
optimizer = torch.optim.AdamW([
    {'params': head_params, 'lr': 1e-3, 'name': 'head'},
    {'params': tail_params, 'lr': 1e-4, 'name': 'head'},
])
# after
optimizer = torch.optim.AdamW([
    {'params': head_params, 'lr': 1e-3, 'name': 'head'},
    {'params': tail_params, 'lr': 1e-4, 'name': 'tail'},
])

names = [g.get('name') for g in optimizer.param_groups if 'name' in g]
assert len(names) == len(set(names)), 'duplicate param-group names in optimizer'

def no_duplicate_group_names(optimizer) -> bool:
    names = [g.get('name') for g in optimizer.param_groups if 'name' in g]
    return len(names) == len(set(names))

When it happens

Trigger: An optimizer like `Adam([{...,'name':'head'}, {...,'name':'head'}])` — two param groups with the same `name` key. Triggered when the monitor inspects schedulers/optimizers at train start (or first logging).

Common situations: Duplicating param groups by mistake when building grouped optimizers; forgetting to set 'name' at all on multiple groups after Lightning's internal name extraction collapses them; refactor copying a group dict without changing the name.

Related errors

• An optimizer should be passed only once to the `setup` metho
• logging_interval should be `step` or `epoch` or `None`.
• Cannot use `LearningRateMonitor` callback with `Trainer` tha
• When `optimizer.step(closure)` is called, the closure should
• Training with multiple optimizers is only supported with man

AI-assisted analysis of Lightning-AI/pytorch-lightning@9fed5c27d2 (2026-08-28). Data as JSON: /api/errors/53d41f34afcbc561. Report an issue: GitHub.