# tinker.types.DimuonParams

## *class* [**tinker.types.DimuonParams**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/optimizer.py#L45)(*[StrictBase](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/_models.py#L76)*)

Hyperparameters for one Dimuon optimizer step.

The caller supplies schedules. Omitted settings use the defaults on each step; previous overrides do not carry forward. Matrix-sign iteration count and normalization epsilon are internal constants.

**Fields:**

- [**type**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/optimizer.py#L54) (*Literal['dimuon']*) – Optimizer family for this step.
- [**learning_rate**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/optimizer.py#L57) (*float*) – Learning rate applied to a direction with Frobenius norm sqrt(rank), with rank correction applied uniformly to all LoRA matrices.
- [**grad_clip_norm**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/optimizer.py#L60) (*float*) – Maximum global gradient norm. Nonpositive values disable gradient clipping.
- [**beta1**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/optimizer.py#L63) (*float*) – Coefficient used for computing running averages of gradients.
- [**damping**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/optimizer.py#L66) (*float*) – Positive damping that smooths the row and column preconditioner.
- [**rank_lr_correction_exponent**](https://github.com/thinking-machines-lab/tinker/blob/main/src/tinker/types/optimizer.py#L69) (*float*) – Exponent for learning-rate correction relative to rank 32. Zero disables rank correction.