Hyperparameters

AutoScientist proposes a complete training recipe before training, then evaluates and adjusts the recipe across iterations. For most runs, keep the recommended values unchanged. AutoScientist is optimized to select these values together and improve them based on measured results.

Changing individual hyperparameters is not recommended. An override can prevent AutoScientist from exploring the configuration it would otherwise select and can reduce training quality. Override a value only when you have a specific experimental or infrastructure constraint.

Recipe fields

• Algorithm selects LoRA or full fine-tuning. Full fine-tuning is available only for supported models.
• Epochs controls how many passes training makes over the dataset.
• LoRA rank (r) controls the capacity of the low-rank matrices. Higher ranks can represent more complex changes but require more memory and compute.
• Alpha scales the LoRA adapter’s contribution to the base model.
• Model name selects the base model used for training.
• Target layers identifies modules such as q_proj, v_proj, or all linear layers where LoRA adapters are inserted.
• Optimizer and schedule settings control parameter updates and how the learning rate changes. For example, a linear schedule decays steadily, while a cosine schedule slows its decay near the end.
• Warmup ratio is the fraction of training steps used to increase the learning rate gradually from zero.
• Gradient clipping caps the gradient norm to reduce exploding gradients and unstable updates.

These are conceptual names from the app. The accepted API keys and value types are model-dependent; use the recommend_hyperparams reference rather than translating labels into guessed field names.

Review in the app

The proposed fields are editable, but the default workflow is to review and accept them:

1. Review the model, algorithm, and generated recipe.
2. Keep the recommended values unless a known constraint requires an override.
3. Inspect the final AutoScientist Config JSON before confirming.
4. Let AutoScientist evaluate and refine the recipe across iterations.
5. After training, compare win rate and diagnostics as described in Interpreting results.

Advanced: Apply API overrides

API overrides are an advanced escape hatch for controlled experiments or hard requirements such as a fixed base model or training method. The SDK accepts model and a hyperparams mapping, but does not expose a separate method for editing a recommendation. training_type is a field inside hyperparams, not a top-level argument.

If an override is necessary, avoid unsupported key names: copy a valid hyperparams object from the current app configuration or construct it from the current create API schema, save it as hyperparams.json, and pass it unchanged:

import json
from pathlib import Path

from adaption import Adaption

client = Adaption()

dataset_id = "dataset_abc123"
hyperparams = json.loads(Path("hyperparams.json").read_text())

run = client.autoscientist.create(
    dataset_id=dataset_id,
    hyperparams={**hyperparams, "training_type": "lora"},
)
print(run.id, run.status)

Unset hyperparameter keys remain under platform control. To enforce a specific model, review the supported models, list the currently available IDs, and pass one through model:

response = client.autoscientist.list_models()
for model in response.models:
    print(model.id)

Prefer leaving model and hyperparams unset so AutoScientist can choose and iterate on them automatically. See Running AutoScientist for the complete request and status workflow.