IFStruct

IFStruct measures whether a model can generate JSON or YAML that satisfies an explicit schema and structural constraints. Pipette evaluates each published quantization with a deterministic validator rather than a judge model.

Configuration

Field Value
Benchmark ID eval_ifstruct_release_v1_0
Dataset scope Full 2,000-prompt release_v1_0 set; no down-selection
Attempts 1 per prompt; 2,000 completions per model configuration
Sampling Temperature 0.6
Published metric Per-prompt pass rate
Benchmark ID
eval_ifstruct_release_v1_0
Dataset scope
Full 2,000-prompt release_v1_0 set; no down-selection
Attempts
1 per prompt; 2,000 completions per model configuration
Sampling
Temperature 0.6
Published metric
Per-prompt pass rate

Dataset and generation

IFStruct is a Liquid AI benchmark. Pipette uses the full 2,000-prompt release set without down-selection and generates one completion per prompt.

Completions are generated by a runner included in the dashboard's evaluation selection scope. See Evaluation Methodology for the shared generation and authority-selection policy.

Scoring

Liquid's ifstruct validator checks each response deterministically in a fixed order: required code block, JSON or YAML parsing, commentary outside the structured value, top-level structure and wrapper key, per-field schema rules, and item count. Schema rules include type, required fields, enum membership, and numeric bounds.

Pipette removes <think> and [THINK] reasoning blocks before validation. Other commentary outside the structured value fails the no-commentary check. A sample passes only when the validator returns no errors.

The scorer reports the overall pass rate, with additional slices by output format, top-level structure, task type, and common error category.

The release data and sample definitions are in the IFStruct dataset. Validation is implemented in the pinned IFStruct scorer.