llm

package
v0.0.113 Latest Latest
Warning

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

 Go to latest Published: Aug 5, 2024 License: MIT Imports: 28 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.GPT3Dot5TurboInstruct,
	Temperature:      0.7,
	MaxTokens:        256,
	TopP:             1,
	PresencePenalty:  0,
	FrequencyPenalty: 0,
	N:                1,
	BestOf:           1,
	Stream:           false,
	MaxRetries:       3,
}
View Source 
var DefaultPenalty = ai21.Penalty{
	Scale:               0,
	ApplyToWhitespaces:  true,
	ApplyToPunctuations: true,
	ApplyToNumbers:      true,
	ApplyToStopwords:    true,
	ApplyToEmojis:       true,
}

DefaultPenalty represents the default penalty options for AI21 text completion.

Functions

func EnforceStopTokens added in v0.0.71 

func EnforceStopTokens(text string, stop []string) string

EnforceStopTokens cuts off the text as soon as any stop words occur.

Types

type AI21 added in v0.0.70 

type AI21 struct {
	schema.Tokenizer
	// contains filtered or unexported fields
}

AI21 is an AI21 LLM model that generates text based on a provided response function.

func NewAI21 added in v0.0.70 

func NewAI21(apiKey string, optFns ...func(o *AI21Options)) (*AI21, error)

NewAI21 creates a new AI21 LLM instance with the provided API key and optional configuration options.

func NewAI21FromClient added in v0.0.70 

func NewAI21FromClient(client AI21Client, optFns ...func(o *AI21Options)) (*AI21, error)

NewAI21FromClient creates a new AI21 LLM instance using the provided AI21 client and optional configuration options.

func (*AI21) Callbacks added in v0.0.70 

func (l *AI21) Callbacks() []schema.Callback

Callbacks returns the registered callbacks of the model.

func (*AI21) Generate added in v0.0.70 

func (l *AI21) 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 (*AI21) InvocationParams added in v0.0.70 

func (l *AI21) InvocationParams() map[string]any

InvocationParams returns the parameters used in the model invocation.

func (*AI21) Type added in v0.0.70 

func (l *AI21) Type() string

Type returns the type of the model.

func (*AI21) Verbose added in v0.0.70 

func (l *AI21) Verbose() bool

Verbose returns the verbosity setting of the model.

type AI21Client added in v0.0.70 

type AI21Client interface {
	CreateCompletion(ctx context.Context, model string, req *ai21.CompleteRequest) (*ai21.CompleteResponse, error)
}

AI21Client is an interface for interacting with the AI21 API.

type AI21Options added in v0.0.70 

type AI21Options struct {
	// CallbackOptions specify options for handling callbacks during text generation.
	*schema.CallbackOptions `map:"-"`
	// Tokenizer represents the tokenizer to be used with the LLM model.
	schema.Tokenizer `map:"-"`

	// Model is the name of the AI21 model to use for text completion.
	Model string `map:"model,omitempty"`

	// Temperature controls the randomness of text generation. Higher values make it more random.
	Temperature float64 `map:"temperature"`

	// MaxTokens sets the maximum number of tokens in the generated text.
	MaxTokens int `map:"maxTokens"`

	// MinTokens sets the minimum number of tokens in the generated text.
	MinTokens int `map:"minTokens"`

	// TopP sets the nucleus sampling probability. Higher values result in more diverse text.
	TopP float64 `map:"topP"`

	// PresencePenalty specifies the penalty for repeating words in generated text.
	PresencePenalty ai21.Penalty `map:"presencePenalty"`

	// CountPenalty specifies the penalty for repeating tokens in generated text.
	CountPenalty ai21.Penalty `map:"countPenalty"`

	// FrequencyPenalty specifies the penalty for generating frequent words.
	FrequencyPenalty ai21.Penalty `map:"frequencyPenalty"`

	// NumResults sets the number of completion results to return.
	NumResults int `map:"numResults"`
}

AI21Options contains options for configuring the AI21 LLM model.

type AzureOpenAI added in v0.0.38 

type AzureOpenAI struct {
	*OpenAI
	// contains filtered or unexported fields
}

func NewAzureOpenAI added in v0.0.38 

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

func (*AzureOpenAI) InvocationParams added in v0.0.55 

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

InvocationParams returns the parameters used in the model invocation.

func (*AzureOpenAI) Type added in v0.0.38 

func (l *AzureOpenAI) Type() string

Type returns the type of the model.

type AzureOpenAIOptions added in v0.0.38 

type AzureOpenAIOptions struct {
	OpenAIOptions
	Deployment string `map:"deployment,omitempty"`
}

type Bedrock added in v0.0.71 

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

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

func NewBedrock added in v0.0.71 

func NewBedrock(client BedrockRuntimeClient, modelID string, optFns ...func(o *BedrockOptions)) (*Bedrock, error)

NewBedrock creates a new instance of the Bedrock LLM model with the provided response function and options.

func NewBedrockAI21 added in v0.0.78 

func NewBedrockAI21(client BedrockRuntimeClient, optFns ...func(o *BedrockAI21Options)) (*Bedrock, error)

func NewBedrockAmazon added in v0.0.79 

func NewBedrockAmazon(client BedrockRuntimeClient, optFns ...func(o *BedrockAmazonOptions)) (*Bedrock, error)

func NewBedrockAnthropic added in v0.0.110 

func NewBedrockAnthropic(client BedrockRuntimeClient, optFns ...func(o *BedrockAnthropicOptions)) (*Bedrock, error)

func NewBedrockCohere added in v0.0.79 

func NewBedrockCohere(client BedrockRuntimeClient, optFns ...func(o *BedrockCohereOptions)) (*Bedrock, error)

func NewBedrockMeta added in v0.0.87 

func NewBedrockMeta(client BedrockRuntimeClient, optFns ...func(o *BedrockMetaOptions)) (*Bedrock, error)

NewBedrockMeta creates a new instance of Bedrock for the "meta" provider.

func NewBedrockMistral added in v0.0.109 

func NewBedrockMistral(client BedrockRuntimeClient, optFns ...func(o *BedrockMistralOptions)) (*Bedrock, error)

func (*Bedrock) Callbacks added in v0.0.71 

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

Callbacks returns the registered callbacks of the model.

func (*Bedrock) Generate added in v0.0.71 

func (l *Bedrock) 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 (*Bedrock) InvocationParams added in v0.0.71 

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

InvocationParams returns the parameters used in the model invocation.

func (*Bedrock) Type added in v0.0.71 

func (l *Bedrock) Type() string

Type returns the type of the model.

func (*Bedrock) Verbose added in v0.0.71 

func (l *Bedrock) Verbose() bool

Verbose returns the verbosity setting of the model.

type BedrockAI21Options added in v0.0.78 

type BedrockAI21Options struct {
	*schema.CallbackOptions `map:"-"`
	schema.Tokenizer        `map:"-"`

	// Model id to use.
	ModelID string `map:"model_id,omitempty"`

	// Temperature controls the randomness of text generation. Higher values make it more random.
	Temperature float64 `map:"temperature"`

	// TopP sets the nucleus sampling probability. Higher values result in more diverse text.
	TopP float64 `map:"topP"`

	// MaxTokens sets the maximum number of tokens in the generated text.
	MaxTokens int `map:"maxTokens"`

	// PresencePenalty specifies the penalty for repeating words in generated text.
	PresencePenalty ai21.Penalty `map:"presencePenalty"`

	// CountPenalty specifies the penalty for repeating tokens in generated text.
	CountPenalty ai21.Penalty `map:"countPenalty"`

	// FrequencyPenalty specifies the penalty for generating frequent words.
	FrequencyPenalty ai21.Penalty `map:"frequencyPenalty"`

	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

type BedrockAmazonOptions added in v0.0.79 

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

	// Model id to use.
	ModelID string `map:"model_id,omitempty"`

	// Temperature controls the randomness of text generation. Higher values make it more random.
	Temperature float64 `json:"temperature"`

	// TopP is the total probability mass of tokens to consider at each step.
	TopP float64 `json:"topP"`

	// MaxTokenCount sets the maximum number of tokens in the generated text.
	MaxTokenCount int `json:"maxTokenCount"`

	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

type BedrockAnthropicOptions added in v0.0.77 

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

	// Model id to use.
	ModelID string `map:"model_id,omitempty"`

	// MaxTokensToSmaple sets the maximum number of tokens in the generated text.
	MaxTokensToSample int `map:"max_tokens_to_sample"`

	// Temperature controls the randomness of text generation. Higher values make it more random.
	Temperature float32 `map:"temperature"`

	// 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 int `map:"top_k"`

	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

type BedrockCohereOptions added in v0.0.79 

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

	// Model id to use.
	ModelID string `map:"model_id,omitempty"`

	// Temperature controls the randomness of text generation. Higher values make it more random.
	Temperature float64 `json:"temperature,omitempty"`

	// P is the total probability mass of tokens to consider at each step.
	P float64 `json:"p,omitempty"`

	// K determines how the model selects tokens for output.
	K float64 `json:"k,omitempty"`

	// MaxTokens sets the maximum number of tokens in the generated text.
	MaxTokens int `json:"max_tokens,omitempty"`

	// ReturnLikelihoods specifies how and if the token likelihoods are returned with the response.
	ReturnLikelihoods ReturnLikelihood `json:"return_likelihoods,omitempty"`

	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

type BedrockInputOutputAdapter added in v0.0.71 

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

BedrockInputOutputAdapter is a helper struct for preparing input and handling output for Bedrock model.

func NewBedrockInputOutputAdapter added in v0.0.71 

func NewBedrockInputOutputAdapter(provider string) *BedrockInputOutputAdapter

NewBedrockInputOutputAdpter creates a new instance of BedrockInputOutputAdpter.

func (*BedrockInputOutputAdapter) PrepareInput added in v0.0.71 

func (bioa *BedrockInputOutputAdapter) PrepareInput(prompt string, modelParams map[string]any) ([]byte, error)

PrepareInput prepares the input for the Bedrock model based on the specified provider.

func (*BedrockInputOutputAdapter) PrepareOutput added in v0.0.71 

func (bioa *BedrockInputOutputAdapter) PrepareOutput(response []byte) (string, error)

PrepareOutput prepares the output for the Bedrock model based on the specified provider.

func (*BedrockInputOutputAdapter) PrepareStreamOutput added in v0.0.96 

func (bioa *BedrockInputOutputAdapter) PrepareStreamOutput(response []byte) (string, error)

PrepareStreamOutput prepares the output for the Bedrock model based on the specified provider.

type BedrockMetaOptions added in v0.0.87 

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

	// Model id to use.
	ModelID string `map:"model_id,omitempty"`

	// Temperature controls the randomness of text generation. Higher values make it more random.
	Temperature float32 `map:"temperature"`

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

	// MaxGenLen specify the maximum number of tokens to use in the generated response.
	MaxGenLen int `map:"max_gen_len"`

	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

BedrockMetaOptions contains options for configuring the Bedrock model with the "meta" provider.

type BedrockMistralOptions added in v0.0.109 

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

	// Model id to use.
	ModelID string `map:"model_id,omitempty"`

	// Temperature controls the randomness of text generation. Higher values make it more random.
	Temperature float32 `map:"temperature"`

	// 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 int `map:"top_k"`

	// MaxTokens sets the maximum number of tokens in the generated text.
	MaxTokens int `json:"max_tokens,omitempty"`

	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

type BedrockOptions added in v0.0.71 

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

	// Model params to use.
	ModelParams map[string]any `map:"model_params,omitempty"`

	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

BedrockOptions contains options for configuring the Bedrock LLM model.

type BedrockRuntimeClient added in v0.0.71 

type BedrockRuntimeClient interface {
	InvokeModel(ctx context.Context, params *bedrockruntime.InvokeModelInput, optFns ...func(*bedrockruntime.Options)) (*bedrockruntime.InvokeModelOutput, error)
	InvokeModelWithResponseStream(ctx context.Context, params *bedrockruntime.InvokeModelWithResponseStreamInput, optFns ...func(*bedrockruntime.Options)) (*bedrockruntime.InvokeModelWithResponseStreamOutput, error)
}

BedrockRuntimeClient is an interface for the Bedrock model runtime client.

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(ctx context.Context, request *cohere.GenerateRequest, opts ...core.RequestOption) (*cohere.Generation, 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 int `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(fakeResultFunc FakeResultFunc, optFns ...func(o *FakeOptions)) *Fake

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

func NewSimpleFake added in v0.0.58 

func NewSimpleFake(resultText string, optFns ...func(o *FakeOptions)) *Fake

NewSimpleFake creates a simple instance of the Fake LLM model with a fixed response for all inputs.

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:"-"`
	LLMType                 string `map:"-"`
}

FakeOptions contains options for configuring the Fake LLM model.

type FakeResultFunc added in v0.0.58 

type FakeResultFunc func(ctx context.Context, prompt string) (*schema.ModelResult, error)

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

type GoogleGenAI added in v0.0.92 

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

GoogleGenAI represents the GoogleGenAI Language Model.

func NewGoogleGenAI added in v0.0.92 

func NewGoogleGenAI(client GoogleGenAIClient, optFns ...func(o *GoogleGenAIOptions)) (*GoogleGenAI, error)

NewGoogleGenAI creates a new instance of the GoogleGenAI Language Model.

func (*GoogleGenAI) Callbacks added in v0.0.92 

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

Callbacks returns the registered callbacks of the model.

func (*GoogleGenAI) Generate added in v0.0.92 

func (l *GoogleGenAI) 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 (*GoogleGenAI) InvocationParams added in v0.0.92 

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

InvocationParams returns the parameters used in the model invocation.

func (*GoogleGenAI) Type added in v0.0.92 

func (l *GoogleGenAI) Type() string

Type returns the type of the model.

func (*GoogleGenAI) Verbose added in v0.0.92 

func (l *GoogleGenAI) Verbose() bool

Verbose returns the verbosity setting of the model.

type GoogleGenAIClient added in v0.0.92 

type GoogleGenAIClient interface {
	GenerateContent(context.Context, *generativelanguagepb.GenerateContentRequest, ...gax.CallOption) (*generativelanguagepb.GenerateContentResponse, error)
	StreamGenerateContent(ctx context.Context, req *generativelanguagepb.GenerateContentRequest, opts ...gax.CallOption) (generativelanguagepb.GenerativeService_StreamGenerateContentClient, error)
	CountTokens(context.Context, *generativelanguagepb.CountTokensRequest, ...gax.CallOption) (*generativelanguagepb.CountTokensResponse, error)
}

GoogleGenAIClient is an interface for the GoogleGenAI model client.

type GoogleGenAIOptions added in v0.0.92 

type GoogleGenAIOptions struct {
	// CallbackOptions specify options for handling callbacks during text generation.
	*schema.CallbackOptions `map:"-"`
	// Tokenizer represents the tokenizer to be used with the LLM model.
	schema.Tokenizer `map:"-"`
	// ModelName is the name of the GoogleGenAI model to use.
	ModelName string `map:"model_name,omitempty"`
	// CandidateCount is the number of candidate generations to consider.
	CandidateCount int32 `map:"candidate_count,omitempty"`
	// MaxOutputTokens is the maximum number of tokens to generate in the output.
	MaxOutputTokens int32 `map:"max_output_tokens,omitempty"`
	// Temperature controls the randomness of the generation. Higher values make the output more random.
	Temperature float32 `map:"temperature,omitempty"`
	// TopP is the nucleus sampling parameter. It controls the cumulative probability of the most likely tokens to sample from.
	TopP float32 `map:"top_p,omitempty"`
	// TopK is the number of top tokens to consider for sampling.
	TopK int32 `map:"top_k,omitempty"`
	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

GoogleGenAIOptions contains options for the GoogleGenAI Language Model.

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"`
	Options                 huggingface.Options
}

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

type Ollama added in v0.0.90 

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

Ollama is a struct representing the Ollama generative model.

func NewOllama added in v0.0.90 

func NewOllama(client OllamaClient, optFns ...func(o *OllamaOptions)) (*Ollama, error)

NewOllama creates a new instance of the Ollama model with the provided client and options.

func (*Ollama) Callbacks added in v0.0.90 

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

Callbacks returns the registered callbacks of the model.

func (*Ollama) Generate added in v0.0.90 

func (l *Ollama) 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 (*Ollama) InvocationParams added in v0.0.90 

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

InvocationParams returns the parameters used in the model invocation.

func (*Ollama) Type added in v0.0.90 

func (l *Ollama) Type() string

Type returns the type of the model.

func (*Ollama) Verbose added in v0.0.90 

func (l *Ollama) Verbose() bool

Verbose returns the verbosity setting of the model.

type OllamaClient added in v0.0.90 

type OllamaClient interface {
	// CreateGeneration produces a single request and response for the Ollama generative model.
	CreateGeneration(ctx context.Context, req *ollama.GenerationRequest) (*ollama.GenerationResponse, error)
	// CreateGenerationStream initiates a streaming request and returns a stream for the Ollama generative model.
	CreateGenerationStream(ctx context.Context, req *ollama.GenerationRequest) (*ollama.GenerationStream, error)
}

OllamaClient is an interface for the Ollama generative model client.

type OllamaOptions added in v0.0.90 

type OllamaOptions struct {
	// CallbackOptions specify options for handling callbacks during text generation.
	*schema.CallbackOptions `map:"-"`
	// Tokenizer represents the tokenizer to be used with the LLM model.
	schema.Tokenizer `map:"-"`
	// ModelName is the name of the Gemini model to use.
	ModelName string `map:"model_name,omitempty"`
	// Temperature controls the randomness of the generation. Higher values make the output more random.
	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 nucleus sampling parameter. It controls the cumulative probability of the most likely tokens to sample from.
	TopP float32 `map:"top_p,omitempty"`
	// TopK is the number of top tokens to consider for sampling.
	TopK int `map:"top_k,omitempty"`
	// PresencePenalty penalizes repeated tokens.
	PresencePenalty float32 `map:"presence_penalty,omitempty"`
	// FrequencyPenalty penalizes repeated tokens according to frequency.
	FrequencyPenalty float32 `map:"frequency_penalty,omitempty"`
	// Stream indicates whether to stream the results or not.
	Stream bool `map:"stream,omitempty"`
}

OllamaOptions contains options for the Ollama 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 `map:"base_url,omitempty"`
	// OrgID is the organization ID for accessing the OpenAI service.
	OrgID string `map:"org_id,omitempty"`
}

OpenAIOptions contains options for configuring the OpenAI LLM model.

type ReturnLikelihood added in v0.0.79 

type ReturnLikelihood string
const (
	ReturnLikelihoodGeneration ReturnLikelihood = "GENERATION"
	ReturnLikelihoodAll        ReturnLikelihood = "ALL"
	ReturnLikelihoodNone       ReturnLikelihood = "NONE"
)

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.