Mengira bilangan Token yang dihantar ke LLM dalam Go (bahagian 2)

pengenalan

Ini ialah bahagian kedua kerja menulis aplikasi Go untuk menentukan bilangan token yang dihantar oleh pengguna kepada LLM berdasarkan teks yang dipilih.

Dalam artikel sebelum ini saya menyebut saya ingin membina sesuatu yang ditulis dalam Golang sahaja, dan antara repositori Github yang saya lihat, yang ini nampaknya sangat bagus: go-hggingface. Kod itu nampaknya sangat baharu tetapi ia "semacam" berfungsi untuk saya.

Perlaksanaan

Pertama sekali, kod tersebut mengakses Hugginface untuk mendapatkan senarai semua "tokenizer" yang disertakan dengan LLM, jadi pengguna harus mempunyai token HF. Jadi saya meletakkan token saya dalam fail .env seperti yang ditunjukkan.

HF_TOKEN="your-huggingface-token"

Kemudian menggunakan contoh yang disediakan dalam halaman berikut (https://github.com/gomlx/go-huggingface?tab=readme-ov-file) saya membina kod saya sendiri di sekelilingnya.

package main

import (
 "bytes"
 "fmt"
 "log"
 "os"
 "os/exec"
 "runtime"

 "github.com/gomlx/go-huggingface/hub"
 "github.com/gomlx/go-huggingface/tokenizers"

 "github.com/joho/godotenv"
 "github.com/sqweek/dialog"

 "fyne.io/fyne/v2"
 "fyne.io/fyne/v2/app"
 "fyne.io/fyne/v2/container"
 "fyne.io/fyne/v2/widget"
 //"github.com/inancgumus/scree"
)

var (
 // Model IDs we use for testing.
 hfModelIDs = []string{
  "ibm-granite/granite-3.1-8b-instruct",
  "meta-llama/Llama-3.3-70B-Instruct",
  "mistralai/Mistral-7B-Instruct-v0.3",
  "google/gemma-2-2b-it",
  "sentence-transformers/all-MiniLM-L6-v2",
  "protectai/deberta-v3-base-zeroshot-v1-onnx",
  "KnightsAnalytics/distilbert-base-uncased-finetuned-sst-2-english",
  "KnightsAnalytics/distilbert-NER",
  "SamLowe/roberta-base-go_emotions-onnx",
 }
)

func runCmd(name string, arg ...string) {
 cmd := exec.Command(name, arg...)
 cmd.Stdout = os.Stdout
 cmd.Run()
}

func ClearTerminal() {
 switch runtime.GOOS {
 case "darwin":
  runCmd("clear")
 case "linux":
  runCmd("clear")
 case "windows":
  runCmd("cmd", "/c", "cls")
 default:
  runCmd("clear")
 }
}

func FileSelectionDialog() string {
 // Open a file dialog box and let the user select a text file
 filePath, err := dialog.File().Filter("Text Files", "txt").Load()
 if err != nil {
  if err.Error() == "Cancelled" {
   fmt.Println("File selection was cancelled.")
  }
  log.Fatalf("Error selecting file: %v", err)
 }

 // Output the selected file name
 fmt.Printf("Selected file: %s\n", filePath)
 return filePath
}

func main() {

 var filePath string

 // read the '.env' file
 err := godotenv.Load()
 if err != nil {
  log.Fatal("Error loading .env file")
 }
 // get the value of the 'HF_TOKEN' environment variable
 hfAuthToken := os.Getenv("HF_TOKEN")
 if hfAuthToken == "" {
  log.Fatal("HF_TOKEN environment variable is not set")
 }

 // to display a list of LLMs to determine the # of tokens later on regarding the given text
 var llm string = ""
 var modelID string = ""
 myApp := app.New()
 myWindow := myApp.NewWindow("Select a LLM in the list")
 items := hfModelIDs
 // Label to display the selected item
 selectedItem := widget.NewLabel("Selected LLM: None")
 // Create a list widget
 list := widget.NewList(
  func() int {
   // Return the number of items in the list
   return len(items)
  },
  func() fyne.CanvasObject {
   // Template for each list item
   return widget.NewLabel("Template")
  },
  func(id widget.ListItemID, obj fyne.CanvasObject) {
   // Update the template with the actual data
   obj.(*widget.Label).SetText(items[id])
  },
 )
 // Handle list item selection
 list.OnSelected = func(id widget.ListItemID) {
  selectedItem.SetText("Selected LLM:" + items[id])
  llm = items[id]
 }

 // Layout with the list and selected item label
 content := container.NewVBox(
  list,
  selectedItem,
 )

 // Set the content of the window
 myWindow.SetContent(content)
 myWindow.Resize(fyne.NewSize(300, 400))
 myWindow.ShowAndRun()
 ClearTerminal()
 fmt.Printf("Selected LLM: %s\n", llm)
 //////

 //List files for the selected model
 for _, modelID := range hfModelIDs {
  if modelID == llm {
   fmt.Printf("\n%s:\n", modelID)
   repo := hub.New(modelID).WithAuth(hfAuthToken)
   for fileName, err := range repo.IterFileNames() {
    if err != nil {
     panic(err)
    }
    fmt.Printf("fileName\t%s\n", fileName)
    fmt.Printf("repo\t%s\n", repo)
    fmt.Printf("modelID\t%s\n", modelID)
   }
  }
 }

 //List tokenizer classes for the selected model
 for _, modelID := range hfModelIDs {
  if modelID == llm {
   fmt.Printf("\n%s:\n", modelID)
   repo := hub.New(modelID).WithAuth(hfAuthToken)
   fmt.Printf("\trepo=%s\n", repo)
   config, err := tokenizers.GetConfig(repo)
   if err != nil {
    panic(err)
   }
   fmt.Printf("\ttokenizer_class=%s\n", config.TokenizerClass)
  }
 }

 // Models URL -> "https://huggingface.co/api/models"
 repo := hub.New(modelID).WithAuth(hfAuthToken)
 tokenizer, err := tokenizers.New(repo)
 if err != nil {
  panic(err)
 }

 // call file selection dialogbox
 filePath = FileSelectionDialog()

 // Open the file
 filerc, err := os.Open(filePath)
 if err != nil {
  fmt.Printf("Error opening file: %v\n", err)
  return
 }
 defer filerc.Close()

 // Put the text file content into a buffer and convert it to a string.
 buf := new(bytes.Buffer)
 buf.ReadFrom(filerc)
 sentence := buf.String()

 tokens := tokenizer.Encode(sentence)
 fmt.Println("Sentence:\n", sentence)

 fmt.Printf("Tokens:  \t%v\n", tokens)
}

Dalam bahagian “var” untuk “hfModelIDs” saya menambahkan beberapa rujukan baharu seperti model Granite, Meta LLama dan juga model Mistral IBM.

Token Huggingface diperoleh secara langsung dan dibaca di dalam kod Go juga.

Saya menambah kotak dialog untuk memaparkan senarai LLM (yang akan saya ubah akhirnya), kotak dialog untuk menambah teks daripada fail (saya suka perkara seperti itu ?) dan beberapa baris kod untuk mengosongkan dan bersihkan skrin?!

Teks input ialah seperti berikut;

The popularity of the Rust language continues to explode; yet, many critical codebases remain authored in C, and cannot be realistically rewritten by hand. Automatically translating C to Rust is thus an appealing course of action. Several works have gone down this path, handling an ever-increasing subset of C through a variety of Rust features, such as unsafe. While the prospect of automation is appealing, producing code that relies on unsafe negates the memory safety guarantees offered by Rust, and therefore the main advantages of porting existing codebases to memory-safe languages.
We instead explore a different path, and explore what it would take to translate C to safe Rust; that is, to produce code that is trivially memory safe, because it abides by Rust's type system without caveats. Our work sports several original contributions: a type-directed translation from (a subset of) C to safe Rust; a novel static analysis based on "split trees" that allows expressing C's pointer arithmetic using Rust's slices and splitting operations; an analysis that infers exactly which borrows need to be mutable; and a compilation strategy for C's struct types that is compatible with Rust's distinction between non-owned and owned allocations.
We apply our methodology to existing formally verified C codebases: the HACL* cryptographic library, and binary parsers and serializers from EverParse, and show that the subset of C we support is sufficient to translate both applications to safe Rust. Our evaluation shows that for the few places that do violate Rust's aliasing discipline, automated, surgical rewrites suffice; and that the few strategic copies we insert have a negligible performance impact. Of particular note, the application of our approach to HACL* results in a 80,000 line verified cryptographic library, written in pure Rust, that implements all modern algorithms - the first of its kind.

Ujian
Kod sebaik sahaja dilaksanakan menunjukkan dialog bx di mana anda boleh memilih LLM yang diingini.

Jika semuanya berjalan lancar, langkah seterusnya ialah memuat turun fail "tokenizer" secara setempat (rujuk penjelasan repo Github) dan kemudian kotak dialog ditunjukkan untuk memilih fail teks dengan kandungan yang akan dinilai dari segi istilah daripada bilangan token.

Setakat ini, saya telah meminta akses kepada model Meta LLama dan Google “google/gemma-2–2b-it” dan sedang menunggu akses untuk diberikan.

google/gemma-2-2b-it:
        repo=google/gemma-2-2b-it
panic: request for metadata from "https://huggingface.co/google/gemma-2-2b-it/resolve/299a8560bedf22ed1c72a8a11e7dce4a7f9f51f8/tokenizer_config.json" failed with the following message: "403 Forbidden"

Kesimpulan

Saya fikir berada di jalan yang betul untuk mencapai apa yang saya ingin dapatkan, program Golang yang dapat menentukan bilangan token ialah pertanyaan pengguna yang dihantar ke LLM.

Satu-satunya matlamat projek ini adalah untuk mempelajari sistem dalaman di sebalik penentuan bilangan token dalam pertanyaan terhadap pelbagai LLM dan untuk mengetahui cara ia dikira.

Terima kasih kerana membaca dan terbuka kepada komen.

Dan sehingga kesimpulan akhir, nantikan… ?

Atas ialah kandungan terperinci Mengira bilangan Token yang dihantar ke LLM dalam Go (bahagian 2). Untuk maklumat lanjut, sila ikut artikel berkaitan lain di laman web China PHP!