vertexai(test): add corpora_test in tokenizer module (#10784)

vertexai/genai/tokenizer/corpora_test.go

@@ -0,0 +1,287 @@
+// Copyright 2024 Google LLC
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//	http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+package tokenizer
+
+import (
+	"archive/zip"
+	"bytes"
+	"context"
+	"fmt"
+	"io"
+	"log"
+	"net/http"
+	"os"
+	"strings"
+	"testing"
+
+	"cloud.google.com/go/vertexai/genai"
+	"golang.org/x/text/encoding"
+	"golang.org/x/text/encoding/charmap"
+	"golang.org/x/text/encoding/japanese"
+	"golang.org/x/text/encoding/simplifiedchinese"
+	"golang.org/x/text/transform"
+)
+
+// corporaInfo holds the name and content of a file in the zip archive
+type corporaInfo struct {
+	Name    string
+	Content []byte
+}
+
+// corporaGenerator is a helper function that downloads a zip archive from a given URL,
+// extracts the content of each file in the archive,
+// and returns a slice of corporaInfo objects containing the name and content of each file.
+func corporaGenerator(url string) ([]corporaInfo, error) {
+	var corpora []corporaInfo
+
+	// Download the zip file
+	resp, err := http.Get(url)
+	if err != nil {
+		return nil, fmt.Errorf("error downloading file: %v", err)
+	}
+	defer resp.Body.Close()
+
+	// Read the content of the response body
+	body, err := io.ReadAll(resp.Body)
+	if err != nil {
+		return nil, fmt.Errorf("error reading response body: %v", err)
+	}
+
+	// Create a zip reader from the downloaded content
+	zipReader, err := zip.NewReader(bytes.NewReader(body), int64(len(body)))
+	if err != nil {
+		return nil, fmt.Errorf("error creating zip reader: %v", err)
+	}
+
+	// Iterate over each file in the zip archive
+	for _, file := range zipReader.File {
+		fileReader, err := file.Open()
+		if err != nil {
+			return nil, fmt.Errorf("error opening file: %v", err)
+		}
+
+		// Check if the file is a text file
+		if !file.FileInfo().IsDir() && file.FileInfo().Mode().IsRegular() {
+			content, err := io.ReadAll(fileReader)
+			fileReader.Close()
+			if err != nil {
+				return nil, fmt.Errorf("error reading file content: %v", err)
+			}
+
+			corpora = append(corpora, corporaInfo{
+				Name:    file.Name[len("udhr/"):],
+				Content: content,
+			})
+		}
+	}
+
+	return corpora, nil
+}
+
+// udhrCorpus represents the Universal Declaration of Human Rights (UDHR) corpus.
+// This corpus contains translations of the UDHR into many languages,
+// stored in a specific directory structure within a zip archive.
+//
+// The files in the corpus USUALLY follow a naming convention:
+//
+//	<Language>_<Script>-<Encoding>
+//
+// For example:
+//   - English_English-UTF8
+//   - French_Français-Latin1
+//   - Spanish_Español-UTF8
+//
+// The Language and Script parts are self-explanatory.
+// The Encoding part indicates the character encoding used in the file.
+//
+// This corpus is used to test the token counting functionality
+// against a diverse set of languages and encodings.
+type udhrCorpus struct {
+	EncodingByFileSuffix map[string]encoding.Encoding
+	EncodingByFilename   map[string]encoding.Encoding
+
+	// Skip lists files that should be skipped during testing.
+	// This is useful for excluding files that are known to cause issues
+	// or are not relevant for the test.
+	Skip map[string]bool
+}
+
+// newUdhrCorpus initializes a new udhrCorpus with encoding patterns and skip set
+// func newUdhrCorpus() *udhrCorpus {
+func newUdhrCorpus() *udhrCorpus {
+
+	EncodingByFileSuffix := map[string]encoding.Encoding{
+		"Latin1":   charmap.ISO8859_1,
+		"Hebrew":   charmap.ISO8859_8,
+		"Arabic":   charmap.Windows1256,
+		"UTF8":     encoding.Nop,
+		"Cyrillic": charmap.Windows1251,
+		"SJIS":     japanese.ShiftJIS,
+		"GB2312":   simplifiedchinese.HZGB2312,
+		"Latin2":   charmap.ISO8859_2,
+		"Greek":    charmap.ISO8859_7,
+		"Turkish":  charmap.ISO8859_9,
+		"Baltic":   charmap.ISO8859_4,
+		"EUC":      japanese.EUCJP,
+		"VPS":      charmap.Windows1258,
+		"Agra":     encoding.Nop,
+		"T61":      charmap.ISO8859_3,
+	}
+
+	// For non-conventional filenames:
+	EncodingByFilename := map[string]encoding.Encoding{
+		"Czech_Cesky-UTF8":              charmap.Windows1250,
+		"Polish-Latin2":                 charmap.Windows1250,
+		"Polish_Polski-Latin2":          charmap.Windows1250,
+		"Amahuaca":                      charmap.ISO8859_1,
+		"Turkish_Turkce-Turkish":        charmap.ISO8859_9,
+		"Lithuanian_Lietuviskai-Baltic": charmap.ISO8859_4,
+		"Abkhaz-Cyrillic+Abkh":          charmap.Windows1251,
+		"Azeri_Azerbaijani_Cyrillic-Az.Times.Cyr.Normal0117": charmap.Windows1251,
+		"Azeri_Azerbaijani_Latin-Az.Times.Lat0117":           charmap.ISO8859_2,
+	}
+
+	// The skip list comes from the NLTK source code which says these are unsupported encodings,
+	// or in general encodings Go doesn't support.
+	// See NLTK source code reference: https://github.com/nltk/nltk/blob/f6567388b4399000b9aa2a6b0db713bff3fe332a/nltk/corpus/reader/udhr.py#L14
+	Skip := map[string]bool{
+		// The following files are not fully decodable because they
+		// were truncated at wrong bytes:
+		"Burmese_Myanmar-UTF8":     true,
+		"Japanese_Nihongo-JIS":     true,
+		"Chinese_Mandarin-HZ":      true,
+		"Chinese_Mandarin-UTF8":    true,
+		"Gujarati-UTF8":            true,
+		"Hungarian_Magyar-Unicode": true,
+		"Lao-UTF8":                 true,
+		"Magahi-UTF8":              true,
+		"Marathi-UTF8":             true,
+		"Tamil-UTF8":               true,
+		"Magahi-Agrarpc":           true,
+		"Magahi-Agra":              true,
+		// encoding not supported in Go.
+		"Vietnamese-VIQR": true,
+		"Vietnamese-TCVN": true,
+		// The following files are encoded for specific fonts:
+		"Burmese_Myanmar-WinResearcher":  true,
+		"Armenian-DallakHelv":            true,
+		"Tigrinya_Tigrigna-VG2Main":      true,
+		"Amharic-Afenegus6..60375":       true,
+		"Navaho_Dine-Navajo-Navaho-font": true,
+		// The following files are unintended:
+		"Czech-Latin2-err":     true,
+		"Russian_Russky-UTF8~": true,
+	}
+
+	return &udhrCorpus{
+		EncodingByFileSuffix: EncodingByFileSuffix,
+		EncodingByFilename:   EncodingByFilename,
+		Skip:                 Skip,
+	}
+}
+
+// getEncoding returns the encoding for a given filename based on patterns
+func (ucr *udhrCorpus) getEncoding(filename string) (encoding.Encoding, bool) {
+	if enc, exists := ucr.EncodingByFilename[filename]; exists {
+		return enc, true
+	}
+
+	parts := strings.Split(filename, "-")
+	encodingKey := parts[len(parts)-1]
+	if enc, exists := ucr.EncodingByFileSuffix[encodingKey]; exists {
+		return enc, true
+	}
+
+	return nil, false
+}
+
+// shouldSkip checks if the file should be skipped
+func (ucr *udhrCorpus) shouldSkip(filename string) bool {
+	return ucr.Skip[filename]
+}
+
+// decodeBytes decodes the given byte slice using the specified encoding
+func decodeBytes(enc encoding.Encoding, content []byte) (string, error) {
+	decodedBytes, _, err := transform.Bytes(enc.NewDecoder(), content)
+	if err != nil {
+		return "", fmt.Errorf("error decoding bytes: %v", err)
+	}
+	return string(decodedBytes), nil
+}
+
+const defaultModel = "gemini-1.0-pro"
+const defaultLocation = "us-central1"
+
+func TestCountTokensWithCorpora(t *testing.T) {
+	projectID := os.Getenv("VERTEX_PROJECT_ID")
+	if testing.Short() {
+		t.Skip("skipping live test in -short mode")
+	}
+
+	if projectID == "" {
+		t.Skip("set a VERTEX_PROJECT_ID env var to run live tests")
+	}
+	ctx := context.Background()
+	client, err := genai.NewClient(ctx, projectID, defaultLocation)
+	if err != nil {
+		t.Fatal(err)
+	}
+	defer client.Close()
+	model := client.GenerativeModel(defaultModel)
+	ucr := newUdhrCorpus()
+
+	corporaURL := "https://raw.githubusercontent.com/nltk/nltk_data/gh-pages/packages/corpora/udhr.zip"
+	files, err := corporaGenerator(corporaURL)
+	if err != nil {
+		t.Fatalf("Failed to generate corpora: %v", err)
+	}
+
+	// Iterate over files generated by the generator function
+	for _, fileInfo := range files {
+		if ucr.shouldSkip(fileInfo.Name) {
+			fmt.Printf("Skipping file: %s\n", fileInfo.Name)
+			continue
+		}
+
+		enc, found := ucr.getEncoding(fileInfo.Name)
+		if !found {
+			fmt.Printf("No encoding found for file: %s\n", fileInfo.Name)
+			continue
+		}
+
+		decodedContent, err := decodeBytes(enc, fileInfo.Content)
+		if err != nil {
+			log.Fatalf("Failed to decode bytes: %v", err)
+		}
+
+		tok, err := New(defaultModel)
+		if err != nil {
+			log.Fatal(err)
+		}
+
+		localNtoks, err := tok.CountTokens(genai.Text(decodedContent))
+		if err != nil {
+			log.Fatal(err)
+		}
+		remoteNtoks, err := model.CountTokens(ctx, genai.Text(decodedContent))
+		if err != nil {
+			log.Fatal(fileInfo.Name, err)
+		}
+		if localNtoks.TotalTokens != remoteNtoks.TotalTokens {
+			t.Errorf("expected %d(remote count-token results), but got %d(local count-token results)", remoteNtoks, localNtoks)
+		}
+
+	}
+
+}

vertexai/go.mod

@@ -6,6 +6,7 @@ require (
 	cloud.google.com/go v0.115.1
 	cloud.google.com/go/aiplatform v1.68.0
 	github.com/google/go-cmp v0.6.0
+	golang.org/x/text v0.17.0
 	google.golang.org/api v0.196.0
 	google.golang.org/genproto v0.0.0-20240903143218-8af14fe29dc1
 	google.golang.org/protobuf v1.34.2
@@ -35,7 +36,6 @@ require (
 	golang.org/x/oauth2 v0.22.0 // indirect
 	golang.org/x/sync v0.8.0 // indirect
 	golang.org/x/sys v0.24.0 // indirect
-	golang.org/x/text v0.17.0 // indirect
 	golang.org/x/time v0.6.0 // indirect
 	google.golang.org/genproto/googleapis/api v0.0.0-20240903143218-8af14fe29dc1 // indirect
 	google.golang.org/genproto/googleapis/rpc v0.0.0-20240903143218-8af14fe29dc1 // indirect