llm

package
v0.0.53 Latest Latest
Warning

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

 Go to latest Published: Jul 27, 2023 License: MIT Imports: 20 Imported by: 0

Documentation

Overview

Package llm provides functionalities for working with Large Language Models (LLMs).

Index

Constants

This section is empty.

Variables

View Source 
var DefaultOpenAIOptions = OpenAIOptions{
	CallbackOptions: &schema.CallbackOptions{
		Verbose: golc.Verbose,
	},
	ModelName:        openai.GPT3TextDavinci003,
	Temperature:      0.7,
	MaxTokens:        256,
	TopP:             1,
	PresencePenalty:  0,
	FrequencyPenalty: 0,
	N:                1,
	BestOf:           1,
	Stream:           false,
	MaxRetries:       3,
}

Functions

This section is empty.

Types

type AzureOpenAI added in v0.0.38 

type AzureOpenAI struct {
	*OpenAI
}

func NewAzureOpenAI added in v0.0.38 

func NewAzureOpenAI(apiKey, baseURL string, optFns ...func(o *AzureOpenAIOptions)) (*AzureOpenAI, error)

func (*AzureOpenAI) Type added in v0.0.38 

func (cm *AzureOpenAI) Type() string

Type returns the type of the model.

type AzureOpenAIOptions added in v0.0.38 

type AzureOpenAIOptions struct {
	OpenAIOptions
	Deployment string
}

type Cohere 

type Cohere struct {
	schema.Tokenizer
	// contains filtered or unexported fields
}

Cohere represents the Cohere language model.

func NewCohere 

func NewCohere(apiKey string, optFns ...func(o *CohereOptions)) (*Cohere, error)

NewCohere creates a new Cohere instance using the provided API key and optional configuration options. It internally creates a Cohere client using the provided API key and initializes the Cohere struct.

func NewCohereFromClient added in v0.0.31 

func NewCohereFromClient(client CohereClient, optFns ...func(o *CohereOptions)) (*Cohere, error)

NewCohereFromClient creates a new Cohere instance using the provided Cohere client and optional configuration options.

func (*Cohere) Callbacks 

func (l *Cohere) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*Cohere) Generate 

func (l *Cohere) Generate(ctx context.Context, prompt string, optFns ...func(o *schema.GenerateOptions)) (*schema.ModelResult, error)

Generate generates text based on the provided prompt and options.

func (*Cohere) InvocationParams added in v0.0.27 

func (l *Cohere) InvocationParams() map[string]any

InvocationParams returns the parameters used in the llm model invocation.

func (*Cohere) Type 

func (l *Cohere) Type() string

Type returns the type of the model.

func (*Cohere) Verbose 

func (l *Cohere) Verbose() bool

Verbose returns the verbosity setting of the model.

type CohereClient added in v0.0.31 

type CohereClient interface {
	Generate(opts cohere.GenerateOptions) (*cohere.GenerateResponse, error)
}

CohereClient is an interface for the Cohere client.

type CohereOptions 

type CohereOptions struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`

	// Model represents the name or identifier of the Cohere language model to use.
	Model string `map:"model,omitempty"`

	// NumGenerations denotes the maximum number of generations that will be returned.                   string
	NumGenerations int `map:"num_generations"`

	// MaxTokens denotes the number of tokens to predict per generation.
	MaxTokens uint `map:"max_tokens"`

	// Temperature is a non-negative float that tunes the degree of randomness in generation.
	Temperature float64 `map:"temperature"`

	// K specifies the number of top most likely tokens to consider for generation at each step.
	K int `map:"k"`

	// P is a probability value between 0.0 and 1.0. It ensures that only the most likely tokens,
	// with a total probability mass of P, are considered for generation at each step.
	P float64 `map:"p"`

	// FrequencyPenalty is used to reduce repetitiveness of generated tokens. A higher value applies
	// a stronger penalty to previously present tokens, proportional to how many times they have
	// already appeared in the prompt or prior generation.
	FrequencyPenalty float64 `map:"frequency_penalty"`

	// PresencePenalty is used to reduce repetitiveness of generated tokens. It applies a penalty
	// equally to all tokens that have already appeared, regardless of their exact frequencies.
	PresencePenalty float64 `map:"presence_penalty"`

	// ReturnLikelihoods specifies whether and how the token likelihoods are returned with the response.
	// It can be set to "GENERATION", "ALL", or "NONE". If "GENERATION" is selected, the token likelihoods
	// will only be provided for generated text. If "ALL" is selected, the token likelihoods will be
	// provided for both the prompt and the generated text.
	ReturnLikelihoods string `map:"return_likelihoods,omitempty"`

	// MaxRetries represents the maximum number of retries to make when generating.
	MaxRetries uint `map:"max_retries,omitempty"`
}

CohereOptions contains options for configuring the Cohere LLM model.

type ContentHandler added in v0.0.41 

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

ContentHandler handles content transformation for the LLM model.

func NewContentHandler added in v0.0.41 

func NewContentHandler(contentType, accept string, transformer Transformer) *ContentHandler

NewContentHandler creates a new ContentHandler instance.

func (*ContentHandler) Accept added in v0.0.41 

func (ch *ContentHandler) Accept() string

Accept returns the accept type of the ContentHandler.

func (*ContentHandler) ContentType added in v0.0.41 

func (ch *ContentHandler) ContentType() string

ContentType returns the content type of the ContentHandler.

func (*ContentHandler) TransformInput added in v0.0.41 

func (ch *ContentHandler) TransformInput(prompt string) ([]byte, error)

TransformInput transforms the input prompt using the ContentHandler's transformer.

func (*ContentHandler) TransformOutput added in v0.0.41 

func (ch *ContentHandler) TransformOutput(output []byte) (string, error)

TransformOutput transforms the output from the LLM model using the ContentHandler's transformer.

type Fake added in v0.0.14 

type Fake struct {
	schema.Tokenizer
	// contains filtered or unexported fields
}

Fake is a fake LLM model that generates text based on a provided response function.

func NewFake added in v0.0.14 

func NewFake(responseFunc FakeResponseFunc, optFns ...func(o *FakeOptions)) *Fake

NewFake creates a new instance of the Fake LLM model with the provided response function and options.

func (*Fake) Callbacks added in v0.0.14 

func (l *Fake) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*Fake) Generate added in v0.0.14 

func (l *Fake) Generate(ctx context.Context, prompt string, optFns ...func(o *schema.GenerateOptions)) (*schema.ModelResult, error)

Generate generates text based on the provided prompt and options.

func (*Fake) InvocationParams added in v0.0.27 

func (l *Fake) InvocationParams() map[string]any

InvocationParams returns the parameters used in the model invocation.

func (*Fake) Type added in v0.0.14 

func (l *Fake) Type() string

Type returns the type of the model.

func (*Fake) Verbose added in v0.0.14 

func (l *Fake) Verbose() bool

Verbose returns the verbosity setting of the model.

type FakeOptions added in v0.0.31 

type FakeOptions struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`
}

FakeOptions contains options for configuring the Fake LLM model.

type FakeResponseFunc added in v0.0.35 

type FakeResponseFunc func(prompt string) string

FakeResponseFunc is a function type for generating fake responses based on a prompt.

type HuggingFaceHub added in v0.0.14 

type HuggingFaceHub struct {
	schema.Tokenizer
	// contains filtered or unexported fields
}

HuggingFaceHub represents the Hugging Face Hub LLM model.

func NewHuggingFaceHub added in v0.0.14 

func NewHuggingFaceHub(apiToken string, optFns ...func(o *HuggingFaceHubOptions)) (*HuggingFaceHub, error)

NewHuggingFaceHub creates a new instance of the HuggingFaceHub model using the provided API token and options.

func NewHuggingFaceHubFromClient added in v0.0.31 

func NewHuggingFaceHubFromClient(client HuggingFaceHubClient, optFns ...func(o *HuggingFaceHubOptions)) (*HuggingFaceHub, error)

NewHuggingFaceHubFromClient creates a new instance of the HuggingFaceHub model using the provided client and options.

func (*HuggingFaceHub) Callbacks added in v0.0.14 

func (l *HuggingFaceHub) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*HuggingFaceHub) Generate added in v0.0.14 

func (l *HuggingFaceHub) Generate(ctx context.Context, prompt string, optFns ...func(o *schema.GenerateOptions)) (*schema.ModelResult, error)

Generate generates text based on the provided prompt and options.

func (*HuggingFaceHub) InvocationParams added in v0.0.27 

func (l *HuggingFaceHub) InvocationParams() map[string]any

InvocationParams returns the parameters used in the model invocation.

func (*HuggingFaceHub) Type added in v0.0.14 

func (l *HuggingFaceHub) Type() string

Type returns the type of the model.

func (*HuggingFaceHub) Verbose added in v0.0.14 

func (l *HuggingFaceHub) Verbose() bool

Verbose returns the verbosity setting of the model.

type HuggingFaceHubClient added in v0.0.31 

type HuggingFaceHubClient interface {
	// TextGeneration performs text generation based on the provided request and returns the response.
	TextGeneration(ctx context.Context, req *huggingface.TextGenerationRequest) (huggingface.TextGenerationResponse, error)

	// Text2TextGeneration performs text-to-text generation based on the provided request and returns the response.
	Text2TextGeneration(ctx context.Context, req *huggingface.Text2TextGenerationRequest) (huggingface.Text2TextGenerationResponse, error)

	// Summarization performs text summarization based on the provided request and returns the response.
	Summarization(ctx context.Context, req *huggingface.SummarizationRequest) (huggingface.SummarizationResponse, error)

	// SetModel sets the model to be used for inference.
	SetModel(model string)
}

HuggingFaceHubClient represents the client for interacting with the Hugging Face Hub service.

type HuggingFaceHubOptions added in v0.0.14 

type HuggingFaceHubOptions struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`
	Model                   string `map:"model,omitempty"`
	Task                    string `map:"task,omitempty"`
}

HuggingFaceHubOptions contains options for configuring the Hugging Face Hub LLM model.

type OpenAI 

type OpenAI struct {
	schema.Tokenizer
	// contains filtered or unexported fields
}

OpenAI is an implementation of the LLM interface for the OpenAI language model.

func NewOpenAI 

func NewOpenAI(apiKey string, optFns ...func(o *OpenAIOptions)) (*OpenAI, error)

NewOpenAI creates a new OpenAI instance with the provided API key and options.

func NewOpenAIFromClient added in v0.0.31 

func NewOpenAIFromClient(client OpenAIClient, optFns ...func(o *OpenAIOptions)) (*OpenAI, error)

NewOpenAIFromClient creates a new OpenAI instance with the provided client and options.

func (*OpenAI) Callbacks 

func (l *OpenAI) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*OpenAI) Generate 

func (l *OpenAI) Generate(ctx context.Context, prompt string, optFns ...func(o *schema.GenerateOptions)) (*schema.ModelResult, error)

Generate generates text based on the provided prompt and options.

func (*OpenAI) InvocationParams added in v0.0.27 

func (l *OpenAI) InvocationParams() map[string]any

InvocationParams returns the parameters used in the model invocation.

func (*OpenAI) Type 

func (l *OpenAI) Type() string

Type returns the type of the model.

func (*OpenAI) Verbose 

func (l *OpenAI) Verbose() bool

Verbose returns the verbosity setting of the model.

type OpenAIClient added in v0.0.31 

type OpenAIClient interface {
	// CreateCompletionStream creates a streaming completion request with the provided completion request.
	// It returns a completion stream for receiving streamed completion responses from the OpenAI API.
	// The `CompletionStream` should be closed after use.
	CreateCompletionStream(ctx context.Context, request openai.CompletionRequest) (stream *openai.CompletionStream, err error)

	// CreateCompletion sends a completion request to the OpenAI API and returns the completion response.
	// It blocks until the response is received from the API.
	CreateCompletion(ctx context.Context, request openai.CompletionRequest) (response openai.CompletionResponse, err error)
}

OpenAIClient represents the interface for interacting with the OpenAI API.

type OpenAIOptions 

type OpenAIOptions struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`
	// ModelName is the name of the OpenAI language model to use.
	ModelName string `map:"model_name,omitempty"`
	// Temperature is the sampling temperature to use during text generation.
	Temperature float32 `map:"temperature,omitempty"`
	// MaxTokens is the maximum number of tokens to generate in the completion.
	MaxTokens int `map:"max_tokens,omitempty"`
	// TopP is the total probability mass of tokens to consider at each step.
	TopP float32 `map:"top_p,omitempty"`
	// PresencePenalty penalizes repeated tokens.
	PresencePenalty float32 `map:"presence_penalty,omitempty"`
	// FrequencyPenalty penalizes repeated tokens according to frequency.
	FrequencyPenalty float32 `map:"frequency_penalty,omitempty"`
	// N is the number of completions to generate for each prompt.
	N int `map:"n,omitempty"`
	// BestOf selects the best completion from multiple completions.
	BestOf int `map:"best_of,omitempty"`
	// LogitBias adjusts the probability of specific tokens being generated.
	LogitBias map[string]int `map:"logit_bias,omitempty"`
	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
	// MaxRetries represents the maximum number of retries to make when generating.
	MaxRetries uint `map:"max_retries,omitempty"`
	// BaseURL is the base URL of the OpenAI service.
	BaseURL string
	// OrgID is the organization ID for accessing the OpenAI service.
	OrgID string
}

OpenAIOptions contains options for configuring the OpenAI LLM model.

type Palm added in v0.0.36 

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

Palm is a struct representing the PALM language model.

func NewPalm added in v0.0.36 

func NewPalm(client PalmClient, optFns ...func(o *PalmOptions)) (*Palm, error)

NewPalm creates a new instance of the PALM language model.

func (*Palm) Callbacks added in v0.0.36 

func (l *Palm) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*Palm) Generate added in v0.0.36 

func (l *Palm) Generate(ctx context.Context, prompt string, optFns ...func(o *schema.GenerateOptions)) (*schema.ModelResult, error)

Generate generates text based on the provided prompt and options.

func (*Palm) InvocationParams added in v0.0.36 

func (l *Palm) InvocationParams() map[string]any

InvocationParams returns the parameters used in the model invocation.

func (*Palm) Type added in v0.0.36 

func (l *Palm) Type() string

Type returns the type of the model.

func (*Palm) Verbose added in v0.0.36 

func (l *Palm) Verbose() bool

Verbose returns the verbosity setting of the model.

type PalmClient added in v0.0.36 

type PalmClient interface {
	GenerateText(ctx context.Context, req *generativelanguagepb.GenerateTextRequest, opts ...gax.CallOption) (*generativelanguagepb.GenerateTextResponse, error)
}

PalmClient is the interface for the PALM client.

type PalmOptions added in v0.0.36 

type PalmOptions struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`

	// ModelName is the name of the Palm language model to use.
	ModelName string `map:"model_name,omitempty"`

	// Temperature is the sampling temperature to use during text generation.
	Temperature float32 `map:"temperature,omitempty"`

	// TopP is the total probability mass of tokens to consider at each step.
	TopP float32 `map:"top_p,omitempty"`

	// TopK determines how the model selects tokens for output.
	TopK int32 `map:"top_k"`

	// MaxOutputTokens specifies the maximum number of output tokens for text generation.
	MaxOutputTokens int32 `map:"max_output_tokens"`

	// CandidateCount specifies the number of candidates to generate during text completion.
	CandidateCount int32 `map:"candidate_count"`
}

PalmOptions is the options struct for the PALM language model.

type SagemakerEndpoint 

type SagemakerEndpoint struct {
	schema.Tokenizer
	// contains filtered or unexported fields
}

SagemakerEndpoint represents an LLM model deployed on AWS SageMaker.

func NewSagemakerEndpoint 

func NewSagemakerEndpoint(client SagemakerRuntimeClient, endpointName string, contenHandler *ContentHandler, optFns ...func(o *SagemakerEndpointOptions)) (*SagemakerEndpoint, error)

NewSagemakerEndpoint creates a new SagemakerEndpoint instance.

func (*SagemakerEndpoint) Callbacks 

func (l *SagemakerEndpoint) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*SagemakerEndpoint) Generate 

func (l *SagemakerEndpoint) Generate(ctx context.Context, prompt string, optFns ...func(o *schema.GenerateOptions)) (*schema.ModelResult, error)

Generate generates text based on the provided prompt and options.

func (*SagemakerEndpoint) InvocationParams added in v0.0.27 

func (l *SagemakerEndpoint) InvocationParams() map[string]any

InvocationParams returns the parameters used in the model invocation.

func (*SagemakerEndpoint) Type 

func (l *SagemakerEndpoint) Type() string

Type returns the type of the model.

func (*SagemakerEndpoint) Verbose 

func (l *SagemakerEndpoint) Verbose() bool

Verbose returns the verbosity setting of the model.

type SagemakerEndpointOptions 

type SagemakerEndpointOptions struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`
}

SagemakerEndpointOptions contains options for configuring the SagemakerEndpoint.

type SagemakerRuntimeClient added in v0.0.34 

type SagemakerRuntimeClient interface {
	// InvokeEndpoint invokes an endpoint in the SageMaker Runtime service with the specified input parameters.
	// It returns the output of the endpoint invocation or an error if the invocation fails.
	InvokeEndpoint(ctx context.Context, params *sagemakerruntime.InvokeEndpointInput, optFns ...func(*sagemakerruntime.Options)) (*sagemakerruntime.InvokeEndpointOutput, error)
}

SagemakerRuntimeClient is an interface that represents the client for interacting with the SageMaker Runtime service.

type Transformer 

type Transformer interface {
	// Transforms the input to a format that model can accept
	// as the request Body. Should return bytes or seekable file
	// like object in the format specified in the content_type
	// request header.
	TransformInput(prompt string) ([]byte, error)

	// Transforms the output from the model to string that
	// the LLM class expects.
	TransformOutput(output []byte) (string, error)
}

Transformer defines the interface for transforming input and output data for the LLM model.

type VertexAI added in v0.0.31 

type VertexAI struct {
	schema.Tokenizer
	// contains filtered or unexported fields
}

VertexAI represents the VertexAI language model.

func NewVertexAI added in v0.0.31 

func NewVertexAI(client VertexAIClient, endpoint string, optFns ...func(o *VertexAIOptions)) (*VertexAI, error)

NewVertexAI creates a new VertexAI instance with the provided client and endpoint.

func (*VertexAI) Callbacks added in v0.0.31 

func (l *VertexAI) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*VertexAI) Generate added in v0.0.31 

func (l *VertexAI) Generate(ctx context.Context, prompt string, optFns ...func(o *schema.GenerateOptions)) (*schema.ModelResult, error)

Generate generates text based on the provided prompt and options.

func (*VertexAI) InvocationParams added in v0.0.31 

func (l *VertexAI) InvocationParams() map[string]any

InvocationParams returns the parameters used in the model invocation.

func (*VertexAI) Type added in v0.0.31 

func (l *VertexAI) Type() string

Type returns the type of the model.

func (*VertexAI) Verbose added in v0.0.31 

func (l *VertexAI) Verbose() bool

Verbose returns the verbosity setting of the model.

type VertexAIClient added in v0.0.31 

type VertexAIClient interface {
	// Predict sends a prediction request to the Vertex AI service.
	// It takes a context, predict request, and optional call options.
	// It returns the predict response or an error if the prediction fails.
	Predict(ctx context.Context, req *aiplatformpb.PredictRequest, opts ...gax.CallOption) (*aiplatformpb.PredictResponse, error)
}

VertexAIClient represents the interface for interacting with Vertex AI.

type VertexAIOptions added in v0.0.31 

type VertexAIOptions struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`

	// Temperature is the sampling temperature to use during text generation.
	Temperature float32 `map:"temperature"`

	// MaxOutputTokens determines the maximum amount of text output from one prompt.
	MaxOutputTokens int `map:"max_output_tokens"`

	// TopP is the total probability mass of tokens to consider at each step.
	TopP float32 `map:"top_p"`

	// TopK determines how the model selects tokens for output.
	TopK int `map:"top_k"`
}

VertexAIOptions contains options for configuring the VertexAI language model.

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.