vertebrae.zero_shot
Training-free zero-shot semantic-alignment benchmark runner.
Classes
One frozen extractor evaluated under a fixed prompt protocol. |
|
Aggregated zero-shot comparison results. |
|
One extractor plus its explicit frozen sample and text endpoints. |
|
Compare text-aligned frozen backbones without a learned task head. |
Functions
|
Render a concise report without conflating alignment and overlap scores. |
Module Contents
- class vertebrae.zero_shot.ZeroShotExtractorResult[source]
One frozen extractor evaluated under a fixed prompt protocol.
- class vertebrae.zero_shot.ZeroShotCandidate[source]
One extractor plus its explicit frozen sample and text endpoints.
- class vertebrae.zero_shot.ZeroShotBenchmark(dataset, extractors, *, sample_branch=None, text_branch=None, zero_shot_config=None, scoring_config=None, cache_config=None, compression_config=None, compression_configs=None, embedding_config=None, memory_config=None, resource_profiling_config=None)[source]
Compare text-aligned frozen backbones without a learned task head.
- Parameters:
extractors (Iterable[Any])
sample_branch (Optional[str])
text_branch (Optional[str])
zero_shot_config (Optional[vertebrae.config.ZeroShotConfig])
scoring_config (Optional[vertebrae.config.OverlapScoringConfig])
cache_config (Optional[vertebrae.config.CacheConfig])
compression_config (Optional[vertebrae.config.EmbeddingCompressionConfig])
compression_configs (Optional[Iterable[vertebrae.config.EmbeddingCompressionConfig]])
embedding_config (Optional[vertebrae.config.EmbeddingConfig])
memory_config (Optional[vertebrae.config.MemoryConfig])
resource_profiling_config (Optional[vertebrae.config.ResourceProfilingConfig])
- vertebrae.zero_shot.render_zero_shot_markdown_report(result)[source]
Render a concise report without conflating alignment and overlap scores.
- Parameters:
result (ZeroShotBenchmarkResult)
- Return type:
str