Expand description
Convenience re-exports for the common uni-xervo surface.
use uni_xervo::prelude::*; brings the runtime, catalog/config types, every
task trait (so their methods are in scope on Arc<dyn …> handles), the
common input/result types, and the host-side scoring helpers into scope.
Provider types stay in crate::provider — import the concrete providers
you register explicitly.
Structs§
- DocExtract
Options - Options for a
DocumentExtractionModel::extractcall. - DocExtract
Result - Per-page result of a document extraction call.
- Embed
Result - Result of a dense-embedding call, carrying the vectors plus optional usage.
- Generation
Options - Sampling and length parameters for text generation.
- Generation
Result - The output of a generation call.
- HeadSet
- Selects which embedding heads a
HybridEmbeddingModelshould populate. - Hybrid
Embed Result - Dense, sparse, and multi-vector embeddings from a single forward pass.
- Message
- A single message in a conversation, containing one or more content blocks.
- Model
Alias Spec - Declarative specification that maps a human-readable alias to a concrete provider and model.
- Model
Runtime - The central runtime that owns registered providers and a catalog of model aliases.
- Model
Runtime Builder - Builder for constructing a
ModelRuntimewith registered providers, a model catalog, and a warmup policy. - Multi
Vector Embed Result - Result of a multi-vector embedding call: per-token vectors for each input.
- Multimodal
Input - A heterogeneous input to a
MultimodalEmbeddingModel. - NlpRequest
- One NLP analysis request: a text plus the heads to populate.
- NlpResult
- Structured output of an NLP analysis call.
- NlpTasks
- Selects which NLP heads a caller wants populated.
- Provider
Capabilities - Advertised capabilities of a
ModelProvider. - Retry
Config - Configuration for exponential-backoff retries on transient inference errors.
- Scored
Doc - A single scored document returned by a
RerankerModel. - Sparse
Embed Result - Result of a sparse-embedding call, carrying one
SparseVectorper input. - Tensor
Batch - Tensor
Spec - Token
Usage - Token counts for a generation request.
- Transcribe
Options - Options for a
TranscriptionModel::transcribecall. - Transcribe
Result - Result of a transcription call.
Enums§
- Audio
Input - An audio input to a transcription or audio embedding model.
- Content
Block - A single block of content within a
Message. - DocOutput
Format - Output format for the in-block
contentstring ofDocBlock. - Image
Input - Image data that can be passed as part of a
ContentBlock. - Modality
- Modality labels reported by
MultimodalEmbeddingModel::supported_modalities. - Model
Task - The kind of inference task a model performs.
- Multimodal
Block - One block of input to a
MultimodalEmbeddingModel. - Provider
Health - Health status reported by a provider.
- Runtime
Error - Unified error type covering configuration, loading, inference, and transport failures.
- Tensor
Value - Warmup
Policy - Controls when a model or provider is initialized during runtime startup.
Traits§
- Audio
Embedding Model - A model that produces dense vector embeddings from audio.
- Document
Extraction Model - A vision-language model that extracts structured blocks from document page images.
- Embedding
Model - A model that produces dense vector embeddings from text.
- Generator
Model - A model that generates text, images, or audio from a conversational message history.
- Hybrid
Embedding Model - A model that produces several embedding heads from one shared forward pass.
- Image
Embedding Model - A model that produces dense vector embeddings from images.
- Model
Info - Metadata common to every loaded model handle.
- Model
Provider - A pluggable backend that knows how to load models for one or more
ModelTasktypes. - Multi
Vector Embedding Model - A model that produces per-token (multi-vector / ColBERT) embeddings from text.
- Multimodal
Embedding Model - A model that produces a single dense vector from heterogeneous content (text + image + audio together).
- NlpModel
- A multi-head structured-NLP model.
- OcrModel
- A model that recognizes text in images (optical character recognition).
- RawTensor
Model - Reranker
Model - A model that re-scores documents against a query for relevance ranking.
- Sparse
Embedding Model - A model that produces learned-sparse term-weight vectors from text.
- Transcription
Model - A model that transcribes speech audio into text with timing information.
Functions§
- colbert_
rerank - Rank
docsagainstquerybymax_sim, returning one score per document. - max_sim
- Compute the MaxSim late-interaction score between query and document tokens.
- sparse_
dot - Compute the dot product of two learned-sparse vectors.
Type Aliases§
- Result
- Convenience alias used throughout the crate.
- Sparse
Vector - A single learned-sparse vector as
(term_id, weight)pairs.