Base

Shared base for DeepEval native image metrics.

BaseDeepEvalImageMetric(name, batch_status_check_interval=DefaultValues.BATCH_STATUS_CHECK_INTERVAL, batch_max_iterations=DefaultValues.BATCH_MAX_ITERATIONS, models=None, fallback_models=None, aggregation_method=DefaultValues.AGGREGATION_METHOD, max_concurrent_judges=None, skip_multimodal_check=False, **kwargs)

Bases: DeepEvalMetricFactory, ABC

Base for DeepEval native image metrics (e.g. coherence, reference, text-to-image).

These metrics carry their own per-image scoring prompts, so the gllm-evals multimodal judge rules (multimodal / multimodal_rules) do not apply and are intentionally not exposed here. They build the native LLMTestCase through the shared DeepEvalMetricFactory adapter, which rewrites [ATTACHMENT:...] placeholders into DeepEval [DEEPEVAL:IMAGE:<id>] references and resolves the images through DeepEvalLLMWrapper at judge invocation.

Because every native image metric requires image content, a row without any resolved image fails here as a hard validation error before judge invocation, rather than reaching DeepEval with a non-multimodal test case.