feat(infra): integrate PaddleOCR's PP-StructureV3 as a document parser backend (#714)

This commit is contained in:
Lin Manhui
2025-08-13 16:37:42 +08:00
committed by GitHub
parent 708a6ed0c0
commit 6b60c07c22
30 changed files with 657 additions and 174 deletions

View File

@@ -254,14 +254,16 @@ func (b *basicServices) toPluginServiceComponents() *plugin.ServiceComponents {
func (b *basicServices) toKnowledgeServiceComponents(memoryService *memory.MemoryApplicationServices) *knowledge.ServiceComponents {
return &knowledge.ServiceComponents{
DB: b.infra.DB,
IDGenSVC: b.infra.IDGenSVC,
Storage: b.infra.TOSClient,
RDB: memoryService.RDBDomainSVC,
ImageX: b.infra.ImageXClient,
ES: b.infra.ESClient,
EventBus: b.eventbus.resourceEventBus,
CacheCli: b.infra.CacheCli,
DB: b.infra.DB,
IDGenSVC: b.infra.IDGenSVC,
Storage: b.infra.TOSClient,
RDB: memoryService.RDBDomainSVC,
ImageX: b.infra.ImageXClient,
ES: b.infra.ESClient,
EventBus: b.eventbus.resourceEventBus,
CacheCli: b.infra.CacheCli,
OCR: b.infra.OCR,
ParserManager: b.infra.ParserManager,
}
}

View File

@@ -19,25 +19,37 @@ package appinfra
import (
"context"
"fmt"
"net/http"
"os"
"strconv"
"strings"
"gorm.io/gorm"
"github.com/volcengine/volc-sdk-golang/service/visual"
"github.com/coze-dev/coze-studio/backend/application/internal"
"github.com/coze-dev/coze-studio/backend/infra/contract/cache"
"github.com/coze-dev/coze-studio/backend/infra/contract/chatmodel"
"github.com/coze-dev/coze-studio/backend/infra/contract/coderunner"
"github.com/coze-dev/coze-studio/backend/infra/contract/document/ocr"
"github.com/coze-dev/coze-studio/backend/infra/contract/document/parser"
"github.com/coze-dev/coze-studio/backend/infra/contract/imagex"
"github.com/coze-dev/coze-studio/backend/infra/contract/modelmgr"
"github.com/coze-dev/coze-studio/backend/infra/impl/cache/redis"
"github.com/coze-dev/coze-studio/backend/infra/impl/coderunner/direct"
"github.com/coze-dev/coze-studio/backend/infra/impl/coderunner/sandbox"
"github.com/coze-dev/coze-studio/backend/infra/impl/document/ocr/ppocr"
"github.com/coze-dev/coze-studio/backend/infra/impl/document/ocr/veocr"
builtinParser "github.com/coze-dev/coze-studio/backend/infra/impl/document/parser/builtin"
"github.com/coze-dev/coze-studio/backend/infra/impl/document/parser/ppstructure"
"github.com/coze-dev/coze-studio/backend/infra/impl/es"
"github.com/coze-dev/coze-studio/backend/infra/impl/eventbus"
"github.com/coze-dev/coze-studio/backend/infra/impl/idgen"
"github.com/coze-dev/coze-studio/backend/infra/impl/imagex/veimagex"
"github.com/coze-dev/coze-studio/backend/infra/impl/mysql"
"github.com/coze-dev/coze-studio/backend/infra/impl/storage"
"github.com/coze-dev/coze-studio/backend/pkg/logs"
"github.com/coze-dev/coze-studio/backend/types/consts"
)
@@ -52,6 +64,8 @@ type AppDependencies struct {
AppEventProducer eventbus.Producer
ModelMgr modelmgr.Manager
CodeRunner coderunner.Runner
OCR ocr.OCR
ParserManager parser.Manager
}
func Init(ctx context.Context) (*AppDependencies, error) {
@@ -102,6 +116,14 @@ func Init(ctx context.Context) (*AppDependencies, error) {
deps.CodeRunner = initCodeRunner()
deps.OCR = initOCR()
imageAnnotationModel, _, err := internal.GetBuiltinChatModel(ctx, "IA_")
if err != nil {
return nil, err
}
deps.ParserManager, err = initParserManager(deps.TOSClient, deps.OCR, imageAnnotationModel)
return deps, nil
}
@@ -183,3 +205,48 @@ func initCodeRunner() coderunner.Runner {
return direct.NewRunner()
}
}
func initOCR() ocr.OCR {
var ocr ocr.OCR
switch os.Getenv(consts.OCRType) {
case "ve":
ocrAK := os.Getenv(consts.VeOCRAK)
ocrSK := os.Getenv(consts.VeOCRSK)
if ocrAK == "" || ocrSK == "" {
logs.Warnf("[ve_ocr] ak / sk not configured, ocr might not work well")
}
inst := visual.NewInstance()
inst.Client.SetAccessKey(ocrAK)
inst.Client.SetSecretKey(ocrSK)
ocr = veocr.NewOCR(&veocr.Config{Client: inst})
case "paddleocr":
url := os.Getenv(consts.PPOCRAPIURL)
client := &http.Client{}
ocr = ppocr.NewOCR(&ppocr.Config{Client: client, URL: url})
default:
// accept ocr not configured
}
return ocr
}
func initParserManager(storage storage.Storage, ocr ocr.OCR, imageAnnotationModel chatmodel.BaseChatModel) (parser.Manager, error) {
var parserManager parser.Manager
parserType := os.Getenv(consts.ParserType)
switch parserType {
case "builtin":
parserManager = builtinParser.NewManager(storage, ocr, imageAnnotationModel)
case "paddleocr":
url := os.Getenv(consts.PPStructureAPIURL)
client := &http.Client{}
apiConfig := &ppstructure.APIConfig{
Client: client,
URL: url,
}
parserManager = ppstructure.NewManager(apiConfig, ocr, storage, imageAnnotationModel)
default:
return nil, fmt.Errorf("unexpected document parser type, type=%s", parserType)
}
return parserManager, nil
}

View File

@@ -20,7 +20,6 @@ import (
"context"
"encoding/json"
"fmt"
netHTTP "net/http"
"os"
"path/filepath"
"strconv"
@@ -33,7 +32,6 @@ import (
"github.com/cloudwego/eino/schema"
"github.com/milvus-io/milvus/client/v2/milvusclient"
"github.com/volcengine/volc-sdk-golang/service/vikingdb"
"github.com/volcengine/volc-sdk-golang/service/visual"
"gorm.io/gorm"
"github.com/coze-dev/coze-studio/backend/application/internal"
@@ -42,6 +40,7 @@ import (
"github.com/coze-dev/coze-studio/backend/infra/contract/cache"
"github.com/coze-dev/coze-studio/backend/infra/contract/document/nl2sql"
"github.com/coze-dev/coze-studio/backend/infra/contract/document/ocr"
"github.com/coze-dev/coze-studio/backend/infra/contract/document/parser"
"github.com/coze-dev/coze-studio/backend/infra/contract/document/searchstore"
"github.com/coze-dev/coze-studio/backend/infra/contract/embedding"
"github.com/coze-dev/coze-studio/backend/infra/contract/es"
@@ -52,8 +51,6 @@ import (
"github.com/coze-dev/coze-studio/backend/infra/contract/storage"
chatmodelImpl "github.com/coze-dev/coze-studio/backend/infra/impl/chatmodel"
builtinNL2SQL "github.com/coze-dev/coze-studio/backend/infra/impl/document/nl2sql/builtin"
"github.com/coze-dev/coze-studio/backend/infra/impl/document/ocr/veocr"
builtinParser "github.com/coze-dev/coze-studio/backend/infra/impl/document/parser/builtin"
"github.com/coze-dev/coze-studio/backend/infra/impl/document/rerank/rrf"
sses "github.com/coze-dev/coze-studio/backend/infra/impl/document/searchstore/elasticsearch"
ssmilvus "github.com/coze-dev/coze-studio/backend/infra/impl/document/searchstore/milvus"
@@ -70,14 +67,16 @@ import (
)
type ServiceComponents struct {
DB *gorm.DB
IDGenSVC idgen.IDGenerator
Storage storage.Storage
RDB rdb.RDB
ImageX imagex.ImageX
ES es.Client
EventBus search.ResourceEventBus
CacheCli cache.Cmdable
DB *gorm.DB
IDGenSVC idgen.IDGenerator
Storage storage.Storage
RDB rdb.RDB
ImageX imagex.ImageX
ES es.Client
EventBus search.ResourceEventBus
CacheCli cache.Cmdable
OCR ocr.OCR
ParserManager parser.Manager
}
func InitService(c *ServiceComponents) (*KnowledgeApplicationService, error) {
@@ -102,26 +101,6 @@ func InitService(c *ServiceComponents) (*KnowledgeApplicationService, error) {
}
sManagers = append(sManagers, mgr)
var ocrImpl ocr.OCR
switch os.Getenv("OCR_TYPE") {
case "ve":
ocrAK := os.Getenv("VE_OCR_AK")
ocrSK := os.Getenv("VE_OCR_SK")
if ocrAK == "" || ocrSK == "" {
logs.Warnf("[ve_ocr] ak / sk not configured, ocr might not work well")
}
inst := visual.NewInstance()
inst.Client.SetAccessKey(ocrAK)
inst.Client.SetSecretKey(ocrSK)
ocrImpl = veocr.NewOCR(&veocr.Config{Client: inst})
case "paddleocr":
ppocrURL := os.Getenv("PADDLEOCR_OCR_API_URL")
client := &netHTTP.Client{}
ocrImpl = veocr.NewPPOCR(&veocr.PPOCRConfig{Client: client, URL: ppocrURL})
default:
// accept ocr not configured
}
root, err := os.Getwd()
if err != nil {
logs.Warnf("[InitConfig] Failed to get current working directory: %v", err)
@@ -158,26 +137,20 @@ func InitService(c *ServiceComponents) (*KnowledgeApplicationService, error) {
}
}
imageAnnoChatModel, configured, err := internal.GetBuiltinChatModel(ctx, "IA_")
if err != nil {
return nil, err
}
knowledgeDomainSVC, knowledgeEventHandler := knowledgeImpl.NewKnowledgeSVC(&knowledgeImpl.KnowledgeSVCConfig{
DB: c.DB,
IDGen: c.IDGenSVC,
RDB: c.RDB,
Producer: knowledgeProducer,
SearchStoreManagers: sManagers,
ParseManager: builtinParser.NewManager(c.Storage, ocrImpl, imageAnnoChatModel), // default builtin
Storage: c.Storage,
Rewriter: rewriter,
Reranker: rrf.NewRRFReranker(0), // default rrf
NL2Sql: n2s,
OCR: ocrImpl,
CacheCli: c.CacheCli,
IsAutoAnnotationSupported: configured,
ModelFactory: chatmodelImpl.NewDefaultFactory(),
DB: c.DB,
IDGen: c.IDGenSVC,
RDB: c.RDB,
Producer: knowledgeProducer,
SearchStoreManagers: sManagers,
ParseManager: c.ParserManager,
Storage: c.Storage,
Rewriter: rewriter,
Reranker: rrf.NewRRFReranker(0), // default rrf
NL2Sql: n2s,
OCR: c.OCR,
CacheCli: c.CacheCli,
ModelFactory: chatmodelImpl.NewDefaultFactory(),
})
if err = eventbus.DefaultSVC().RegisterConsumer(nameServer, consts.RMQTopicKnowledge, consts.RMQConsumeGroupKnowledge, knowledgeEventHandler); err != nil {