inference

package
v0.1.0 Latest Latest
Warning

This package is not in the latest version of its module.

 Go to latest Published: Sep 2, 2026 License: Apache-2.0 Imports: 17 Imported by: 0

Documentation

Overview

Package inference defines provider-neutral generation, embedding, token, and usage contracts.

Index

Constants

This section is empty.

Variables

This section is empty.

Functions

func ProbeTextModel 

func ProbeTextModel(ctx context.Context, model TextModel) error

ProbeTextModel sends the smallest provider-neutral generation request. It intentionally bypasses structured generation, tracing, usage reporting, and prompt assets; the caller owns the wall-clock timeout and result cache.

Types

type BatchedEmbeddingModel 

type BatchedEmbeddingModel struct {
	// contains filtered or unexported fields
}

func NewBatchedEmbeddingModel 

func NewBatchedEmbeddingModel(
	transport EmbeddingTransport,
	profile EmbeddingProfile,
	batchSize int,
	limits *Limits,
) (*BatchedEmbeddingModel, error)

func (*BatchedEmbeddingModel) Embed 

func (m *BatchedEmbeddingModel) Embed(ctx context.Context, texts []string) (EmbeddingResult, error)

func (*BatchedEmbeddingModel) Profile 

func (m *BatchedEmbeddingModel) Profile() EmbeddingProfile

type ConfigurationError 

type ConfigurationError struct {
	// contains filtered or unexported fields
}

ConfigurationError reports a stable, non-transient inference configuration failure. A wrapped provider cause is available only through errors.Is/As and is never interpolated into the stable Error value.

func NewConfigurationError 

func NewConfigurationError(code, detail string) *ConfigurationError

func WrapConfigurationError 

func WrapConfigurationError(code, detail string, cause error) *ConfigurationError

WrapConfigurationError retains the provider or SDK error for errors.Is/As without allowing its potentially sensitive text into the stable Error value.

func (*ConfigurationError) Code 

func (e *ConfigurationError) Code() string

func (*ConfigurationError) Detail 

func (e *ConfigurationError) Detail() string

func (*ConfigurationError) Error 

func (e *ConfigurationError) Error() string

func (*ConfigurationError) Unwrap 

func (e *ConfigurationError) Unwrap() error

type EmbeddingInputType 

type EmbeddingInputType string
const (
	EmbeddingDocument EmbeddingInputType = "document"
	EmbeddingQuery    EmbeddingInputType = "query"
)

type EmbeddingModel 

type EmbeddingModel interface {
	Profile() EmbeddingProfile
	Embed(context.Context, []string) (EmbeddingResult, error)
}

type EmbeddingProfile 

type EmbeddingProfile interface {
	ID() string
	ModelName() string
	DimensionCount() int
	NormalizationMode() string
}

type EmbeddingRequest 

type EmbeddingRequest struct {
	// contains filtered or unexported fields
}

func NewEmbeddingRequest 

func NewEmbeddingRequest(inputs []string, inputType EmbeddingInputType, dimension int) (EmbeddingRequest, error)

func (EmbeddingRequest) DimensionCount 

func (r EmbeddingRequest) DimensionCount() int

func (EmbeddingRequest) InputType 

func (r EmbeddingRequest) InputType() EmbeddingInputType

func (EmbeddingRequest) Inputs 

func (r EmbeddingRequest) Inputs() []string

type EmbeddingResult 

type EmbeddingResult struct {
	Vectors [][]float64
	Usage   Usage
}

type EmbeddingTransport 

type EmbeddingTransport interface {
	Embed(context.Context, EmbeddingRequest) (ProviderEmbeddingResult, error)
}

func TraceEmbeddingTransport 

func TraceEmbeddingTransport(transport EmbeddingTransport, provider trace.TracerProvider) EmbeddingTransport

TraceEmbeddingTransport instruments each provider batch while leaving the batching, validation, normalization, and total timeout under BatchedEmbeddingModel's ownership.

type GenerationResult 

type GenerationResult[T any] struct {
	Output T
	Usage  Usage
}

type GenerationSettings 

type GenerationSettings struct {
	// contains filtered or unexported fields
}

func NewGenerationSettings 

func NewGenerationSettings(temperature *float64, maxTokens *int64) (GenerationSettings, error)

func (GenerationSettings) MaxTokens 

func (s GenerationSettings) MaxTokens() *int64

func (GenerationSettings) Temperature 

func (s GenerationSettings) Temperature() *float64

type InvalidOutputError 

type InvalidOutputError struct {
	// contains filtered or unexported fields
}

InvalidOutputError reports a provider result that violated the capability contract. Detail is intentionally bounded and must not contain provider response bodies, prompts, or user content.

func NewInvalidOutputError 

func NewInvalidOutputError(operation, detail string) *InvalidOutputError

func (*InvalidOutputError) Detail 

func (e *InvalidOutputError) Detail() string

func (*InvalidOutputError) Error 

func (e *InvalidOutputError) Error() string

func (*InvalidOutputError) Operation 

func (e *InvalidOutputError) Operation() string

type JSONCodec 

type JSONCodec[I, O any] struct {
	// contains filtered or unexported fields
}

func NewJSONCodec 

func NewJSONCodec[I, O any](
	schema []byte,
	encode func(I) ([]byte, error),
	decode func([]byte) (O, error),
) (*JSONCodec[I, O], error)

func (*JSONCodec[I, O]) DecodeOutput 

func (c *JSONCodec[I, O]) DecodeOutput(value []byte) (O, error)

func (*JSONCodec[I, O]) EncodeInput 

func (c *JSONCodec[I, O]) EncodeInput(value I) ([]byte, error)

func (*JSONCodec[I, O]) OutputSchema 

func (c *JSONCodec[I, O]) OutputSchema() []byte

type Limits 

type Limits struct {
	// contains filtered or unexported fields
}

func DefaultLimits 

func DefaultLimits() Limits

func NewLimits 

func NewLimits(timeout time.Duration, maxRequests int) (Limits, error)

func (Limits) MaxRequests 

func (l Limits) MaxRequests() int

func (Limits) Timeout 

func (l Limits) Timeout() time.Duration

type Message 

type Message struct {
	// contains filtered or unexported fields
}

func NewMessage 

func NewMessage(role MessageRole, content string) (Message, error)

func (Message) Content 

func (m Message) Content() string

func (Message) ResponseIdentity 

func (m Message) ResponseIdentity() *ResponseIdentity

func (Message) Role 

func (m Message) Role() MessageRole

type MessageRole 

type MessageRole uint8
const (
	RoleUser MessageRole = iota + 1
	RoleAssistant
)

type PromptedGenerator 

type PromptedGenerator[I, O any] struct {
	// contains filtered or unexported fields
}

func NewPromptedGenerator 

func NewPromptedGenerator[I, O any](
	model TextModel,
	instructions string,
	codec StructuredCodec[I, O],
	limits *Limits,
	settings GenerationSettings,
) (*PromptedGenerator[I, O], error)

func (*PromptedGenerator[I, O]) Generate 

func (g *PromptedGenerator[I, O]) Generate(ctx context.Context, value I) (GenerationResult[O], error)

type ProviderEmbeddingResult 

type ProviderEmbeddingResult struct {
	// contains filtered or unexported fields
}

func NewProviderEmbeddingResult 

func NewProviderEmbeddingResult(
	inputs []string,
	inputType EmbeddingInputType,
	embeddings [][]float64,
	usage Usage,
) (ProviderEmbeddingResult, error)

func (ProviderEmbeddingResult) Embeddings 

func (r ProviderEmbeddingResult) Embeddings() [][]float64

func (ProviderEmbeddingResult) InputType 

func (r ProviderEmbeddingResult) InputType() EmbeddingInputType

func (ProviderEmbeddingResult) Inputs 

func (r ProviderEmbeddingResult) Inputs() []string

func (ProviderEmbeddingResult) Usage 

func (r ProviderEmbeddingResult) Usage() Usage

type ResponseIdentity 

type ResponseIdentity struct {
	// contains filtered or unexported fields
}

ResponseIdentity carries the minimum provider-generated identity needed to replay an assistant turn without retaining an entire provider response.

func NewResponseIdentity 

func NewResponseIdentity(protocol, itemID string) (ResponseIdentity, error)

func (ResponseIdentity) ItemID 

func (i ResponseIdentity) ItemID() string

func (ResponseIdentity) Protocol 

func (i ResponseIdentity) Protocol() string

type StructuredCodec 

type StructuredCodec[I, O any] interface {
	EncodeInput(I) ([]byte, error)
	OutputSchema() []byte
	DecodeOutput([]byte) (O, error)
}

StructuredCodec keeps wire JSON and validation rules separate from domain values. Implementations must never return partially validated output.

type StructuredGenerator 

type StructuredGenerator[I, O any] interface {
	Generate(context.Context, I) (GenerationResult[O], error)
}

type TextModel 

type TextModel interface {
	Complete(context.Context, TextRequest) (TextResponse, error)
}

TextModel is the deliberately narrow provider contract consumed by the PromptedOutput-compatible structured generator.

func TraceTextModel 

func TraceTextModel(model TextModel, provider trace.TracerProvider) TextModel

TraceTextModel instruments one physical provider request. The span contract deliberately contains no prompt, message, output, model request parameters, credentials, or provider response bodies.

type TextRequest 

type TextRequest struct {
	// contains filtered or unexported fields
}

func (TextRequest) Instructions 

func (r TextRequest) Instructions() []string

func (TextRequest) Messages 

func (r TextRequest) Messages() []Message

func (TextRequest) Settings 

func (r TextRequest) Settings() GenerationSettings

func (TextRequest) StructuredOutput 

func (r TextRequest) StructuredOutput() bool

type TextResponse 

type TextResponse struct {
	// contains filtered or unexported fields
}

func NewTextResponse 

func NewTextResponse(content string, usage Usage) (TextResponse, error)

func (TextResponse) Content 

func (r TextResponse) Content() string

func (TextResponse) ResponseIdentity 

func (r TextResponse) ResponseIdentity() *ResponseIdentity

func (TextResponse) Usage 

func (r TextResponse) Usage() Usage

func (TextResponse) WithResponseIdentity 

func (r TextResponse) WithResponseIdentity(identity ResponseIdentity) TextResponse

type TimeoutError 

type TimeoutError struct {
	// contains filtered or unexported fields
}

TimeoutError reports that the PowerContext wall-clock budget was exhausted.

func NewTimeoutError 

func NewTimeoutError(operation string, timeout time.Duration) *TimeoutError

func (*TimeoutError) Error 

func (e *TimeoutError) Error() string

func (*TimeoutError) Operation 

func (e *TimeoutError) Operation() string

func (*TimeoutError) Timeout 

func (e *TimeoutError) Timeout() time.Duration

type TokenEstimator 

type TokenEstimator struct {
	// contains filtered or unexported fields
}

TokenEstimator validates the result of a deployment-fixed token counter.

func CharacterTokenEstimator 

func CharacterTokenEstimator() *TokenEstimator

func NewTokenEstimator 

func NewTokenEstimator(profile TokenEstimatorProfile, count func(string) int) (*TokenEstimator, error)

func (*TokenEstimator) Estimate 

func (e *TokenEstimator) Estimate(text string) (int, error)

func (*TokenEstimator) Profile 

func (e *TokenEstimator) Profile() TokenEstimatorProfile

type TokenEstimatorProfile 

type TokenEstimatorProfile struct {
	// contains filtered or unexported fields
}

TokenEstimatorProfile is the stable deployment identity of a token policy.

func NewTokenEstimatorProfile 

func NewTokenEstimatorProfile(estimatorID, version string) (TokenEstimatorProfile, error)

func (TokenEstimatorProfile) EstimatorID 

func (p TokenEstimatorProfile) EstimatorID() string

func (TokenEstimatorProfile) Version 

func (p TokenEstimatorProfile) Version() string

type UnavailableError 

type UnavailableError struct {
	// contains filtered or unexported fields
}

UnavailableError is a transient provider or network failure.

func NewUnavailableError 

func NewUnavailableError(operation string) *UnavailableError

func WrapUnavailableError 

func WrapUnavailableError(operation string, cause error) *UnavailableError

WrapUnavailableError retains the provider or network error for errors.Is/As while keeping provider response bodies out of the stable Error value.

func (*UnavailableError) Error 

func (e *UnavailableError) Error() string

func (*UnavailableError) Operation 

func (e *UnavailableError) Operation() string

func (*UnavailableError) Unwrap 

func (e *UnavailableError) Unwrap() error

type Usage 

type Usage struct {
	Requests     int
	InputTokens  *int64
	OutputTokens *int64
}

Usage contains portable successful provider usage fields.

type ValidationError 

type ValidationError struct {
	// contains filtered or unexported fields
}

func NewValidationError 

func NewValidationError(issues []ValidationIssue) (*ValidationError, error)

func (*ValidationError) Error 

func (e *ValidationError) Error() string

func (*ValidationError) Issues 

func (e *ValidationError) Issues() []ValidationIssue

type ValidationIssue 

type ValidationIssue struct {
	// contains filtered or unexported fields
}

func NewValidationIssue 

func NewValidationIssue(kind string, location []any, message string, input json.RawMessage) (ValidationIssue, error)

Source Files

View all Source files

Jump to

Keyboard shortcuts

? : This menu
/ : Search site
f or F : Jump to
y or Y : Canonical URL
go.dev uses cookies from Google to deliver and enhance the quality of its services and to analyze traffic. Learn more.