EvaluateModelRequestSchema

eval_typestring[]required

Array of benchmark evaluation types to run. Available options include mix_eval (comprehensive mix), math, gsm8k, mmlu, arc, hellaswag, drop, triviaqa, and others. Pass ["mix_eval"] to run the standard comprehensive benchmark suite.

Possible values: >= 1

Example: ["mix_eval"]

model_namestringrequired

Name of the deployed fine-tuned model to evaluate.

Example: example-ftmodel

base_modelstring

Base model identifier to evaluate. Must be a fine-tunable model name.

Default: null

benchmarks_to_keepstring[]

Subset of benchmark names to retain in the evaluation results. If empty, all benchmarks are returned.

Default: null

data_percentageinteger

Percentage of the evaluation dataset to sample. Range: 1–100. Lower values speed up evaluation at the cost of statistical precision.

Possible values: >= 1 and <= 100

Default: null

EvaluateModelRequestSchema

{
  "model_name": "test-model-123 ",
  "eval_type": [
    "mix_eval"
  ]
}