# tinker_cookbook.distillation.sdft.Config

## *class* [**tinker_cookbook.distillation.sdft.Config**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L736)()

Configuration for SDFT training.

**Key parameters:**

- `topk`: Number of top tokens for distillation (default 20). Set to 0 for the importance-sampling fallback. K=20 matches full-vocabulary KL in practice.
- `learning_rate`: For LoRA, use 5e-4 to 1e-3. The top-K CE loss produces larger gradients than SFT at the same LR due to more completion tokens per step (on-policy generation), so use the lower end of the range.
- `teacher_sync_every`: Optional periodic hard-sync of student weights into the teacher (approximating EMA). `None` = static frozen teacher, which works comparably to EMA in our experiments.

See `main` for the training loop.

**Fields:**

- [**model_name**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L756) (*str*) – Model
- [**recipe_name**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L757) (*str*)
- [**renderer_name**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L758) (*str | None*, default: `None`)
- [**lora_rank**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L759) (*int*, default: `128`)
- [**base_url**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L760) (*str | None*, default: `None`)
- [**learning_rate**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L763) (*float*, default: `2e-05`) – Training
- [**max_tokens**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L764) (*int*, default: `2048`)
- [**temperature**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L765) (*float*, default: `1.0`)
- [**loss_fn**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L766) (*LossFnType*, default: `'cross_entropy'`)
- [**topk**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L769) (*int*, default: `20`) – SDFT-specific
- [**reverse**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L770) (*bool*, default: `False`)
- [**demo_template**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L771) (*str*, default: `DEFAULT_DEMO_TEMPLATE`)
- [**system_prompt**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L772) (*str | None*, default: `None`)
- [**teacher_sync_every**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L773) (*int | None*, default: `None`)
- [**max_context_length**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L774) (*int*, default: `32768`)
- [**evaluator_builders**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L777) (*list[SamplingClientEvaluatorBuilder]*, default: `[]`) – Evaluation
- [**eval_every**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L778) (*int*, default: `20`)
- [**save_every**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L779) (*int*, default: `20`)
- [**num_substeps**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L782) (*int*, default: `1`) – Standard infra
- [**log_path**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L783) (*str*)
- [**wandb_project**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L784) (*str | None*, default: `None`)
- [**wandb_name**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L785) (*str | None*, default: `None`)
- [**load_checkpoint_path**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L786) (*str | None*, default: `None`)
- [**max_steps**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L787) (*int | None*, default: `None`)
- [**enable_trace**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L789) (*bool*, default: `False`)
- [**span_chart_every**](https://github.com/thinking-machines-lab/tinker-cookbook/blob/main/tinker_cookbook/distillation/sdft.py#L790) (*int*, default: `0`)