tiktoken

tiktoken-go

OpenAI's tiktoken in Go, ported for use at Enterpret.

Tiktoken is a fast BPE tokeniser for use with OpenAI's models.

This is a port of the original tiktoken.

Usage

Install

go get github.com/aavaz-ai/tiktoken-go

Cache

Tiktoken-go has the same cache mechanism as the original Tiktoken library.

You can set the cache directory by using the environment variable TIKTOKEN_CACHE_DIR.

Once this variable is set, tiktoken-go will use this directory to cache the token dictionary.

If you don't set this environment variable, tiktoken-go will download the dictionary each time you initialize an encoding for the first time.

Example

get token by encoding

package main

import (
    "fmt"
    "github.com/aavaz-ai/tiktoken-go"
)

func main() (num_tokens int) {
    text = "Hello, world!"
    encoding = "r50k_base"

	tke, err := tiktoken.GetEncoding(encoding)
	if err != nil {
		err = fmt.Errorf("getEncoding: %v", err)
		return
	}

    // encode
	token := tke.Encode(text, nil, nil)

    // num_tokens
    num_tokens = len(token)
}

get token by Model

package main

import (
    "fmt"
    "github.com/aavaz-ai/tiktoken-go"
)

func main() (num_tokens int) {
    text = "Hello, world!"
    encoding = "davinci"

   tkm, err := tiktoken.EncodingForModel(model)
	if err != nil {
		err = fmt.Errorf("getEncoding: %v", err)
		return
	}

	 // encode
	token := tke.Encode(text, nil, nil)

    // num_tokens
    num_tokens = len(token)
}

counting tokens for chat API calls

Below is an example function for counting tokens for messages passed to gpt-3.5-turbo-0301 or gpt-4-0314.

The following code was written by @nasa1024 based on openai-cookbook examples.

Please note that the token calculation method for the message may change at any time, so this code may not necessarily be applicable in the future.

If you need accurate calculation, please refer to the official documentation.

If you find that this code is no longer applicable, please feel free to submit a PR or Issue.

package main

import (
	"fmt"

	"github.com/aavaz-ai/tiktoken-go"
	"github.com/sashabaranov/go-openai"
)

func NumTokensFromMessages(messages []openai.ChatCompletionMessage, model string) (num_tokens int) {
	tkm, err := tiktoken.EncodingForModel(model)
	if err != nil {
		err = fmt.Errorf("EncodingForModel: %v", err)
		fmt.Println(err)
		return
	}

	var tokens_per_message int
	var tokens_per_name int
	if model == "gpt-3.5-turbo-0301" || model == "gpt-3.5-turbo" {
		tokens_per_message = 4
		tokens_per_name = -1
	} else if model == "gpt-4-0314" || model == "gpt-4" {
		tokens_per_message = 3
		tokens_per_name = 1
	} else {
		fmt.Println("Warning: model not found. Using cl100k_base encoding.")
		tokens_per_message = 3
		tokens_per_name = 1
	}

	for _, message := range messages {
		num_tokens += tokens_per_message
		num_tokens += len(tkm.Encode(message.Content, nil, nil))
		num_tokens += len(tkm.Encode(message.Role, nil, nil))
		if message.Name != "" {
			num_tokens += tokens_per_name
		}
	}
	num_tokens += 3
	return num_tokens
}

available encodings

available models