---
title: "Recommended hyperparameters for one model"
method: POST
path: "/api/finetuning-jobs/model-defaults/"
tags: ["finetuning-jobs"]
---

# Recommended hyperparameters for one model

`POST /api/finetuning-jobs/model-defaults/`

## Request

- Base URL: `https://api.overmindlab.ai`
- URL: `https://api.overmindlab.ai/api/finetuning-jobs/model-defaults/`
- Auth: one of:
  - HTTP bearer
  - API key in header `X-Api-Key`
  - API key in cookie `sessionid`
  - API key in header `X-Api-Key`
  - HTTP bearer

Dataset-size-aware defaults (batch, epochs, learning rate, LoRA config) plus per-knob provenance for an arbitrary catalog model — the same derivation as the recommendation rows, no LLM call. Serves the wizard's 'Add model' and model-switch flows so the client never fabricates hyperparameter defaults.

## Request body

- FinetuningModelDefaultsRequestRequest
  - `dataset_id` string, uuid, required
  - `base_model` string, required

## Response `200`

- FinetuningExperiment: ``match`` and ``skill_scores`` are standing among the ``match_pool`` graded candidates for this dataset; ``grade`` and below are percentiles over every model the benchmark artifact tracks.
  - `tier` string, required
  - `model` string, required
  - `display_name` string, required
  - `params` string, required
  - `total_params_b` number, double
  - `context_length_sft` integer, nullable
  - `max_batch_size` integer, nullable
  - `min_batch_size` integer, nullable
  - `match` number, double, nullable, required
  - `match_rank` integer, nullable, required
  - `match_pool` integer, required
  - `grade` number, double, nullable, required
  - `adjusted_grade` number, double, nullable, required
  - `lower_bound` number, double, nullable, required
  - `confidence` 'high' | 'medium' | 'low' | 'none', required: * `high` - high * `medium` - medium * `low` - low * `none` - none
  - `n_benchmarks` integer, required
  - `skill_scores` FinetuningSkillScore[], required
    - `skill` string, required
    - `weight` number, double, required
    - `percentile_in_field` number, double, required
    - `rank_in_field` integer, required
    - `field_n` integer, required
    - `percentile_global` number, double, required
  - `evidence` FinetuningEvidence[], required
    - `benchmark` string, required
    - `skill` string, required
    - `percentile` number, double, required
    - `cohort_n` integer, required
    - `source` string, required
    - `url` string, required
    - `provenance` 'measured' | 'lab_claimed', required: * `measured` - measured * `lab_claimed` - lab_claimed
  - `selected` boolean, required
  - `hyperparams` unknown, required
  - `use_lora` boolean, required
  - `training_type` FinetuningCatalogTrainingType: models.json ``finetuning.training_type`` — which FT methods are allowed.
    - `lora` FinetuningTrainingMethod, required
      - `enabled` boolean, required
      - `context_length` integer, nullable
      - `validated_context_length` boolean
    - `full` FinetuningTrainingMethod, required
      - `enabled` boolean, required
      - `context_length` integer, nullable
      - `validated_context_length` boolean
  - `gpu_config` FinetuningGpuConfig, required
    - `gpu_type` string, nullable, required
    - `num_gpus` integer, required
    - `max_model_len` integer, nullable, required
  - `cost_estimate` FinetuningCostEstimate, required
    - `usd` number, double, required
    - `price_per_million_usd` number, double, nullable, required
    - `trained_tokens` integer, required
    - `minimum_applied` boolean, required
  - `time_estimate` FinetuningTimeEstimate, required
    - `seconds` integer, nullable, required
    - `human` string, required
  - `learning_rate_lora` number, double, required
  - `learning_rate_full` number, double, required
  - `hyperparam_reasons` unknown, required
  - `serving_context` ServingContextPlan
    - `input_tokens` integer, required
    - `output_tokens` integer, required
    - `rows` integer, required
    - `required_context` integer, required
    - `max_model_len` integer, required
    - `model_context_limit` integer, required
    - `estimated` boolean, required
    - `warnings` string[]

---

[API](https://skmtc.dev/overmindlab/apis/overmind-api.md) - [All operations](https://skmtc.dev/overmindlab/apis/overmind-api/llms.txt) - [OpenAPI document](https://skmtc.dev/overmindlab/apis/overmind-api/revisions/0ecfd39e80c6?raw)