## Recommend hyperparameters

**post** `/api/v1/autoscientist/recommend-hyperparams`

Returns platform-default hyperparameters sized to a specific model and your dataset's effective training size (rows + any augmentation you plan to add). Pure computation — nothing is launched. POST /autoscientist applies these optimized values automatically when `hyperparams` is omitted; only pass overrides when strictly necessary.

### Body Parameters

- `dataset_id: string`

  The dataset the hyperparameters are sized for. Must be a dataset you own that has finished processing.

- `model: string`

  Base model id from GET /training-models to tune for. Required — the recommendation is sized to this model and your dataset.

- `augmentation_domain_rows: optional number`

  Number of domain-targeted augmentation rows you plan to add. Defaults to 0. Counted toward the effective training size so the recommendation matches what you will launch.

- `augmentation_general_rows: optional number`

  Number of general-diversity augmentation rows you plan to add. Defaults to 0. Counted toward the effective training size.

### Returns

- `hyperparams: object { base_model_size, batch_size, learning_rate, 14 more }`

  Recommended hyperparameters, sized to your model and effective dataset size. AutoScientist optimizes these values automatically when `hyperparams` is omitted from POST /autoscientist; only pass overrides when strictly necessary.

  - `base_model_size: optional string`

  - `batch_size: optional "max" or number`

    Either Together's literal 'max' or a positive integer.

    - `"max"`

      - `"max"`

    - `number`

  - `learning_rate: optional number`

  - `lora_alpha: optional number`

    Must be 1× or 2× lora_r. Cross-field check is skipped when lora_r is not supplied alongside (the merged-config path uses generateConfig defaults).

  - `lora_dropout: optional number`

  - `lora_r: optional number`

  - `lora_trainable_modules: optional string`

    'all-linear' or comma-separated module list (e.g. 'q_proj,v_proj').

  - `lr_scheduler_type: optional "linear" or "cosine" or "constant"`

    - `"linear"`

    - `"cosine"`

    - `"constant"`

  - `max_grad_norm: optional number`

  - `min_lr_ratio: optional number`

  - `n_epochs: optional number`

  - `n_evals: optional number`

  - `scheduler_num_cycles: optional number`

    HyperparamService.generateConfig defaults to 0.5 (half-cycle cosine decay). Range floor is relaxed to 0 to keep that default valid.

  - `train_on_inputs: optional boolean`

  - `training_type: optional "lora" or "full"`

    Training strategy override. AutoScientist optimizes this by default; set it only when strictly necessary.

    - `"lora"`

    - `"full"`

  - `warmup_ratio: optional number`

  - `weight_decay: optional number`

### Example

```http
curl https://api.prod.adaptionlabs.ai/api/v1/autoscientist/recommend-hyperparams \
    -H 'Content-Type: application/json' \
    -H "Authorization: Bearer $ADAPTION_API_KEY" \
    -d '{
          "dataset_id": "dataset_id",
          "model": "model"
        }'
```

#### Response

```json
{
  "hyperparams": {
    "base_model_size": "8B",
    "batch_size": "max",
    "learning_rate": 0.00005,
    "lora_alpha": 0,
    "lora_dropout": 0,
    "lora_r": 8,
    "lora_trainable_modules": "q_proj,v_proj",
    "lr_scheduler_type": "linear",
    "max_grad_norm": 2,
    "min_lr_ratio": 0.1,
    "n_epochs": 1,
    "n_evals": 5,
    "scheduler_num_cycles": 0.5,
    "train_on_inputs": true,
    "training_type": "lora",
    "warmup_ratio": 0.03,
    "weight_decay": 0
  }
}
```