gpt

package module
v0.0.3 Latest Latest
Warning

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

 Go to latest Published: May 16, 2024 License: MIT Imports: 17 Imported by: 0

README

GoGPT

This is a high level helper library for OpenAI API. It is designed to be easy to use and to provide a simple interface for interacting with the OpenAI API.

I use this in several projects where I need to batch request asynchronous.

usage

Scheduling batched
func main() {
    token := os.Getenv("OPENAI_API_KEY")
	g, err := gpt.NewGpt(token)
    if err != nil {
        log.Fatal(err)
    }

    var toTranslate []string = loadSnippetsToTranslate()
    
    applicationName := "my-translator"
    systemPrompt := `You are a translater translating snippets of text for an ecommerce shop to polish. Answer with a JSON object with the key "translation" and the value being the translated text.`

    currentSnippets := make([]string, 0)
    batches := make(map[string][]string)
    flush := func() {
        batchId, err := session.CreateBatch()
        if err != nil {
            log.Fatal(err)
        }
        batches[batchId] = currentSnippets
        currentSnippets = make([]string, 0)
        session := g.NewBatchSession()
    }

    session := g.NewBatchSession()
    for _, snippet := range toTranslate {
        reqId := fmt.Sprintf("%s-%d", applicationName, time.Now().UnixNano())
        err := session.AddToBatch(reqId, systemPrompt, snippet)
        if errors.Is(err, gpt.ErrExceedsFileLimit) {
            flush()
        }
        if err != nil {
            log.Fatal(err)
        }
        currentSnippets = append(currentSnippets, snippet)
    }
    if len(currentSnippets) > 0 {
        flush()
    }

    // somewhere store the snippets associated with batchId and lineIdx (idx of snippet in the batch)
}
Retrieve a batch

func main() {
    token := os.Getenv("OPENAI_API_KEY")
	g, err := gpt.NewGpt(token)
    if err != nil {
        log.Fatal(err)
    }

    var scheduledSnippets map[string][]string = loadScheduledSnippets()

    for batchId, snippets := range scheduledSnippets {
        for idx, snippet := range snippets {
    		rawResponse, err := session.RetrieveBatchedRequest(ctx, batchId, idx)
            if errors.Is(err, gpt.ErrBatchNotCompleted) {
			    continue
		    }
            if err != nil {
                log.Fatal(err)
            }
            
            var response struct {
			    Translation []string `json:"translation"`
		    }
            err = json.Unmarshal(rawResponse, &response)
            if err != nil {
                log.Fatal(err)
            }
            fmt.Printf("Snippet: %s -> %s\n", snippet, response.Translation)
        }
    }
}

Documentation

Index

Constants

This section is empty.

Variables

View Source 
var (
	ErrRequestBatch = goerror.New("gpt:retrieve_batch", "Failed to retrieve batch")
	ErrCancelBatch  = goerror.New("gpt:cancel_batch", "Failed to cancel batch")
	ErrCreateBatch  = goerror.New("gpt:create_batch", "Failed to create batch")
)
View Source 
var (
	ErrBatchFailed           = goerror.New("gpt:batch_failed", "Batch failed")
	ErrBatchNotCompleted     = goerror.New("gpt:batch_not_completed", "Batch not completed yet")
	ErrSerializeBatchRequest = goerror.New("gpt:serialize_batch_request", "Failed to serialize batch request")
	ErrExceedsFileLimit      = goerror.New("gpt:exceeds_file_limit", "Exceeds file limit")
	ErrParseBatchLine        = goerror.New("gpt:parse_batch_line", "failed to parse batch line")
)
View Source 
var (
	ErrRequestFile = goerror.New("gpt:request_file", "failed to request files")
	ErrDeleteFile  = goerror.New("gpt:delete_file", "Failed to delete file")
	ErrCreateFile  = goerror.New("gpt:create_file", "Failed to create file")
)
View Source 
var ErrGptAsk = goerror.New("gpt_ask", "Error while asking GPT")
View Source 
var ErrInvalidContent = goerror.New("invalid_content", "Invalid content")
View Source 
var ErrTooManyRetries = goerror.New("too_many_retries", "too many retries")
View Source 
var WithCacheDir = func(cacheDir string) Option { return func(g *Gpt) { g.cacheDir = cacheDir } }
View Source 
var WithModel = func(model string) Option {
	return func(g *Gpt) {
		if model != "" {
			g.model = model
		}
	}
}

Functions

func FilterSlice 

func FilterSlice[I any](slice []I, f func(I) bool) []I

func MapSlice 

func MapSlice[I, O any](slice []I, f func(I) O) []O

func RequestToCurl 

func RequestToCurl(req *http.Request) string

func SliceContains 

func SliceContains[I comparable](slice []I, value I) bool

Types

type BackoffRoundTripper 

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

BackoffRoundTripper is a custom http.RoundTripper that limits the number of concurrent requests and applies backoff when receiving a 429 response.

func NewBackoffRoundTripper 

func NewBackoffRoundTripper(transport http.RoundTripper) *BackoffRoundTripper

func (*BackoffRoundTripper) RoundTrip 

func (rt *BackoffRoundTripper) RoundTrip(req *http.Request) (res *http.Response, err error)

type Gpt 

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

func NewGpt 

func NewGpt(token string, opts ...Option) (*Gpt, error)

func (*Gpt) CancelBatch 

func (g *Gpt) CancelBatch(ctx context.Context, batchId string) goerror.TraceableError

func (*Gpt) DeleteFile 

func (g *Gpt) DeleteFile(ctx context.Context, fileId string) goerror.TraceableError

func (*Gpt) NewBatchSession 

func (g *Gpt) NewBatchSession() *GptBatchSession

func (*Gpt) RetrieveBatch 

func (g *Gpt) RetrieveBatch(ctx context.Context, batchId string) (GptBatchResponse, goerror.TraceableError)

func (*Gpt) RetrieveBatches 

func (g *Gpt) RetrieveBatches(ctx context.Context, stati ...GptBatchStatus) ([]GptBatchResponse, goerror.TraceableError)

func (*Gpt) RetrieveFile 

func (g *Gpt) RetrieveFile(ctx context.Context, fileId string) (GptFileResponse, goerror.TraceableError)

func (*Gpt) RetrieveFileContent 

func (g *Gpt) RetrieveFileContent(ctx context.Context, fileId string) ([]byte, goerror.TraceableError)

func (*Gpt) RetrieveFiles 

func (g *Gpt) RetrieveFiles(ctx context.Context) ([]GptFileResponse, goerror.TraceableError)

type GptBatchResponse 

type GptBatchResponse struct {
	ID       string `json:"id"`
	Object   string `json:"object"`
	Endpoint string `json:"endpoint"`
	Errors   *struct {
		Object string          `json:"object"`
		Data   []gptBatchError `json:"data"`
	} `json:"errors"` // Using *string to allow null value
	InputFileID      string         `json:"input_file_id"`
	CompletionWindow string         `json:"completion_window"`
	Status           GptBatchStatus `json:"status"`
	OutputFileID     *string        `json:"output_file_id"`
	ErrorFileID      *string        `json:"error_file_id"`
	CreatedAt        int64          `json:"created_at"`
	InProgressAt     *int64         `json:"in_progress_at"`
	ExpiresAt        *int64         `json:"expires_at"`
	FinalizingAt     *int64         `json:"finalizing_at"`
	CompletedAt      *int64         `json:"completed_at"`
	FailedAt         *int64         `json:"failed_at"` // Using *int64 to allow null value
	ExpiredAt        *int64         `json:"expired_at"`
	CancellingAt     *int64         `json:"cancelling_at"`
	CancelledAt      *int64         `json:"cancelled_at"`
	RequestCounts    struct {
		Total     int `json:"total"`
		Completed int `json:"completed"`
		Failed    int `json:"failed"`
	} `json:"request_counts"`
	Metadata struct {
		CustomerID       string `json:"customer_id"`
		BatchDescription string `json:"batch_description"`
	} `json:"metadata"`
}

type GptBatchSession 

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

func (*GptBatchSession) AddToBatch 

func (s *GptBatchSession) AddToBatch(customRequestId, systemPrompt, userPrompt string) goerror.TraceableError

AddToBatch adds a request to the current batch data. The customRequestId is used to identify the request in the batch. It should have an application wide prefix to avoid collisions with other applications that batch data. Further the customRequestId should be unique, a timestamp is a good choice. The systemPrompt should describe the task and the userPrompt should contain the input data. The callee is responsible for store the lineIdx of the request. If AddToBatch is called the third time, the lineIdx of the request within the current batch is 2. If the batch data exceeds the 512MB limit, ErrExceedsFileLimit is returned, signaling that the s.CreateBatch() should be called to flush the current batch data

func (*GptBatchSession) CreateBatch 

func (s *GptBatchSession) CreateBatch(ctx context.Context, batchName string) (string, goerror.TraceableError)

CreateBatch creates a new batch with the current batch data. The batchName is used to identify the batch. It is the prefix for the file created and the batch created. the batchname should be unique to this application, to differentiate between different batches of different applications. CreateBatch will not clear its data. Create a new session to start a new batch.

func (*GptBatchSession) RetrieveBatchedRequest 

func (s *GptBatchSession) RetrieveBatchedRequest(ctx context.Context, batchId string, lineIdx int) ([]byte, goerror.TraceableError)

RetrieveBatchedRequest retrieves a single request from a batch. The batchId is the id of the batch to retrieve the request from. The lineIdx is the index of the request in the batch. If the batch is not completed yet, ErrBatchNotCompleted is returned, which is more a flag indicating that the request should be retried later. If the batch failed, ErrBatchFailed is returned, which contains the error that caused the batch to fail. RetrieveBatchedRequest returns the raw []byte of the answer GPT gave (respnse.Body.Choices[0].Message.Content), since it is agnostic to the response format (could be JSON, could be plain text). Sometimes GPT messes up, and a JSONL line is malformed. In this case ErrParseBatchLine is returned. If you want to see the file itself causing that, just add a WithCacheDir to GPT instance and the file will be stored in the cache directory.

type GptBatchStatus 

type GptBatchStatus string
const (
	BatchStatusInProgress GptBatchStatus = "in_progress"
	BatchStatusComplete   GptBatchStatus = "completed"
	BatchStatusFailed     GptBatchStatus = "failed"
)

type GptFilePurpose 

type GptFilePurpose string
const (
	Assistants        GptFilePurpose = "assistants"
	Assistants_output GptFilePurpose = "assistants_output"
	Batch             GptFilePurpose = "batch"
	Batch_output      GptFilePurpose = "batch_output"
	FineTune          GptFilePurpose = "fine-tune"
	FineTuneResults   GptFilePurpose = "fine-tune-results"
	Vision            GptFilePurpose = "vision"
)

type GptFileResponse 

type GptFileResponse struct {
	ID        string         `json:"id"`
	Object    string         `json:"object"`
	Bytes     int            `json:"bytes"`
	CreatedAt int64          `json:"created_at"`
	Filename  string         `json:"filename"`
	Purpose   GptFilePurpose `json:"purpose"`
	Status    *GptFileStatus `json:"status,omitempty"`
}

type GptFileStatus 

type GptFileStatus string
const (
	Uploaded  GptFileStatus = "uploaded"
	Processed GptFileStatus = "processed"
	Error     GptFileStatus = "error"
)

type GzipRoundTripper 

type GzipRoundTripper struct {
	Transport http.RoundTripper
}

GzipRoundTripper wraps an http.RoundTripper, adding gzip compression support.

func (*GzipRoundTripper) RoundTrip 

func (g *GzipRoundTripper) RoundTrip(req *http.Request) (*http.Response, error)

RoundTrip executes a single HTTP transaction and handles gzip compression.

type HeaderRoundTripper 

type HeaderRoundTripper struct {
	Transport http.RoundTripper
	// contains filtered or unexported fields
}

HeaderRoundTripper wraps an http.RoundTripper, adding specific headers

func (*HeaderRoundTripper) RoundTrip 

func (h *HeaderRoundTripper) RoundTrip(req *http.Request) (*http.Response, error)

RoundTrip executes a single HTTP transaction and adds custom headers

type Option 

type Option func(*Gpt)

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.