Merge branch 'langgenius:main' into main

Author: ai408

.github/workflows/style.yml

@@ -82,6 +82,33 @@ jobs:
         if: steps.changed-files.outputs.any_changed == 'true'
         run: yarn run lint
 
+  docker-compose-template:
+    name: Docker Compose Template
+    runs-on: ubuntu-latest
+
+    steps:
+      - name: Checkout code
+        uses: actions/checkout@v4
+
+      - name: Check changed files
+        id: changed-files
+        uses: tj-actions/changed-files@v45
+        with:
+          files: |
+            docker/generate_docker_compose
+            docker/.env.example
+            docker/docker-compose-template.yaml
+            docker/docker-compose.yaml
+
+      - name: Generate Docker Compose
+        if: steps.changed-files.outputs.any_changed == 'true'
+        run: |
+          cd docker
+          ./generate_docker_compose
+
+      - name: Check for changes
+        if: steps.changed-files.outputs.any_changed == 'true'
+        run: git diff --exit-code
 
   superlinter:
     name: SuperLinter

api/configs/middleware/vdb/milvus_config.py

@@ -33,3 +33,9 @@ class MilvusConfig(BaseSettings):
         description="Name of the Milvus database to connect to (default is 'default')",
         default="default",
     )
+
+    MILVUS_ENABLE_HYBRID_SEARCH: bool = Field(
+        description="Enable hybrid search features (requires Milvus >= 2.5.0). Set to false for compatibility with "
+        "older versions",
+        default=True,
+    )

api/controllers/console/datasets/datasets.py

@@ -52,12 +52,12 @@ def get(self):
         # provider = request.args.get("provider", default="vendor")
         search = request.args.get("keyword", default=None, type=str)
         tag_ids = request.args.getlist("tag_ids")
-
+        include_all = request.args.get("include_all", default="false").lower() == "true"
         if ids:
             datasets, total = DatasetService.get_datasets_by_ids(ids, current_user.current_tenant_id)
         else:
             datasets, total = DatasetService.get_datasets(
-                page, limit, current_user.current_tenant_id, current_user, search, tag_ids
+                page, limit, current_user.current_tenant_id, current_user, search, tag_ids, include_all
             )
 
         # check embedding setting
@@ -640,6 +640,7 @@ def get(self):
                 | VectorType.MYSCALE
                 | VectorType.ORACLE
                 | VectorType.ELASTICSEARCH
+                | VectorType.ELASTICSEARCH_JA
                 | VectorType.PGVECTOR
                 | VectorType.TIDB_ON_QDRANT
                 | VectorType.LINDORM
@@ -683,6 +684,7 @@ def get(self, vector_type):
                 | VectorType.MYSCALE
                 | VectorType.ORACLE
                 | VectorType.ELASTICSEARCH
+                | VectorType.ELASTICSEARCH_JA
                 | VectorType.COUCHBASE
                 | VectorType.PGVECTOR
                 | VectorType.LINDORM

api/controllers/console/datasets/datasets_document.py

@@ -257,7 +257,8 @@ def post(self, dataset_id):
         parser.add_argument("original_document_id", type=str, required=False, location="json")
         parser.add_argument("doc_form", type=str, default="text_model", required=False, nullable=False, location="json")
         parser.add_argument("retrieval_model", type=dict, required=False, nullable=False, location="json")
-
+        parser.add_argument("embedding_model", type=str, required=False, nullable=True, location="json")
+        parser.add_argument("embedding_model_provider", type=str, required=False, nullable=True, location="json")
         parser.add_argument(
             "doc_language", type=str, default="English", required=False, nullable=False, location="json"
         )

api/controllers/service_api/__init__.py

@@ -7,4 +7,4 @@
 
 from . import index
 from .app import app, audio, completion, conversation, file, message, workflow
-from .dataset import dataset, document, hit_testing, segment
+from .dataset import dataset, document, hit_testing, segment, upload_file

api/controllers/service_api/dataset/dataset.py

@@ -31,8 +31,11 @@ def get(self, tenant_id):
         # provider = request.args.get("provider", default="vendor")
         search = request.args.get("keyword", default=None, type=str)
         tag_ids = request.args.getlist("tag_ids")
+        include_all = request.args.get("include_all", default="false").lower() == "true"
 
-        datasets, total = DatasetService.get_datasets(page, limit, tenant_id, current_user, search, tag_ids)
+        datasets, total = DatasetService.get_datasets(
+            page, limit, tenant_id, current_user, search, tag_ids, include_all
+        )
         # check embedding setting
         provider_manager = ProviderManager()
         configurations = provider_manager.get_configurations(tenant_id=current_user.current_tenant_id)

api/controllers/service_api/dataset/upload_file.py

@@ -0,0 +1,54 @@
+from werkzeug.exceptions import NotFound
+
+from controllers.service_api import api
+from controllers.service_api.wraps import (
+    DatasetApiResource,
+)
+from core.file import helpers as file_helpers
+from extensions.ext_database import db
+from models.dataset import Dataset
+from models.model import UploadFile
+from services.dataset_service import DocumentService
+
+
+class UploadFileApi(DatasetApiResource):
+    def get(self, tenant_id, dataset_id, document_id):
+        """Get upload file."""
+        # check dataset
+        dataset_id = str(dataset_id)
+        tenant_id = str(tenant_id)
+        dataset = db.session.query(Dataset).filter(Dataset.tenant_id == tenant_id, Dataset.id == dataset_id).first()
+        if not dataset:
+            raise NotFound("Dataset not found.")
+        # check document
+        document_id = str(document_id)
+        document = DocumentService.get_document(dataset.id, document_id)
+        if not document:
+            raise NotFound("Document not found.")
+        # check upload file
+        if document.data_source_type != "upload_file":
+            raise ValueError(f"Document data source type ({document.data_source_type}) is not upload_file.")
+        data_source_info = document.data_source_info_dict
+        if data_source_info and "upload_file_id" in data_source_info:
+            file_id = data_source_info["upload_file_id"]
+            upload_file = db.session.query(UploadFile).filter(UploadFile.id == file_id).first()
+            if not upload_file:
+                raise NotFound("UploadFile not found.")
+        else:
+            raise ValueError("Upload file id not found in document data source info.")
+
+        url = file_helpers.get_signed_file_url(upload_file_id=upload_file.id)
+        return {
+            "id": upload_file.id,
+            "name": upload_file.name,
+            "size": upload_file.size,
+            "extension": upload_file.extension,
+            "url": url,
+            "download_url": f"{url}&as_attachment=true",
+            "mime_type": upload_file.mime_type,
+            "created_by": upload_file.created_by,
+            "created_at": upload_file.created_at.timestamp(),
+        }, 200
+
+
+api.add_resource(UploadFileApi, "/datasets/<uuid:dataset_id>/documents/<uuid:document_id>/upload-file")

api/core/indexing_runner.py

@@ -530,7 +530,6 @@ def _load(
         # chunk nodes by chunk size
         indexing_start_at = time.perf_counter()
         tokens = 0
-        chunk_size = 10
         if dataset_document.doc_form != IndexType.PARENT_CHILD_INDEX:
             # create keyword index
             create_keyword_thread = threading.Thread(
@@ -539,11 +538,22 @@ def _load(
             )
             create_keyword_thread.start()
 
+        max_workers = 10
         if dataset.indexing_technique == "high_quality":
-            with concurrent.futures.ThreadPoolExecutor(max_workers=10) as executor:
+            with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as executor:
                 futures = []
-                for i in range(0, len(documents), chunk_size):
-                    chunk_documents = documents[i : i + chunk_size]
+
+                # Distribute documents into multiple groups based on the hash values of page_content
+                # This is done to prevent multiple threads from processing the same document,
+                # Thereby avoiding potential database insertion deadlocks
+                document_groups: list[list[Document]] = [[] for _ in range(max_workers)]
+                for document in documents:
+                    hash = helper.generate_text_hash(document.page_content)
+                    group_index = int(hash, 16) % max_workers
+                    document_groups[group_index].append(document)
+                for chunk_documents in document_groups:
+                    if len(chunk_documents) == 0:
+                        continue
                     futures.append(
                         executor.submit(
                             self._process_chunk,

api/core/model_runtime/model_providers/__base/tokenizers/gpt2_tokenzier.py

@@ -1,7 +1,8 @@
+import logging
 from threading import Lock
 from typing import Any
 
-import tiktoken
+logger = logging.getLogger(__name__)
 
 _tokenizer: Any = None
 _lock = Lock()
@@ -33,9 +34,18 @@ def get_encoder() -> Any:
             if _tokenizer is None:
                 # Try to use tiktoken to get the tokenizer because it is faster
                 #
-                _tokenizer = tiktoken.get_encoding("gpt2")
-                # base_path = abspath(__file__)
-                # gpt2_tokenizer_path = join(dirname(base_path), "gpt2")
-                # _tokenizer = TransformerGPT2Tokenizer.from_pretrained(gpt2_tokenizer_path)
+                try:
+                    import tiktoken
+
+                    _tokenizer = tiktoken.get_encoding("gpt2")
+                except Exception:
+                    from os.path import abspath, dirname, join
+
+                    from transformers import GPT2Tokenizer as TransformerGPT2Tokenizer  # type: ignore
+
+                    base_path = abspath(__file__)
+                    gpt2_tokenizer_path = join(dirname(base_path), "gpt2")
+                    _tokenizer = TransformerGPT2Tokenizer.from_pretrained(gpt2_tokenizer_path)
+                    logger.info("Fallback to Transformers' GPT-2 tokenizer from tiktoken")
 
             return _tokenizer

api/core/model_runtime/model_providers/openai_api_compatible/llm/llm.py

@@ -377,10 +377,7 @@ def _generate(
                 for tool in tools:
                     formatted_tools.append(helper.dump_model(PromptMessageFunction(function=tool)))
 
-                if prompt_messages[-1].role.value == "tool":
-                    data["tools"] = None
-                else:
-                    data["tools"] = formatted_tools
+                data["tools"] = formatted_tools
 
         if stop:
             data["stop"] = stop

api/core/model_runtime/model_providers/openrouter/llm/claude-3-5-sonnet.yaml

@@ -7,6 +7,7 @@ features:
   - vision
   - tool-call
   - stream-tool-call
+  - document
 model_properties:
   mode: chat
   context_size: 200000

api/core/rag/datasource/vdb/elasticsearch/elasticsearch_ja_vector.py

@@ -0,0 +1,104 @@
+import json
+import logging
+from typing import Any, Optional
+
+from flask import current_app
+
+from core.rag.datasource.vdb.elasticsearch.elasticsearch_vector import (
+    ElasticSearchConfig,
+    ElasticSearchVector,
+    ElasticSearchVectorFactory,
+)
+from core.rag.datasource.vdb.field import Field
+from core.rag.datasource.vdb.vector_type import VectorType
+from core.rag.embedding.embedding_base import Embeddings
+from extensions.ext_redis import redis_client
+from models.dataset import Dataset
+
+logger = logging.getLogger(__name__)
+
+
+class ElasticSearchJaVector(ElasticSearchVector):
+    def create_collection(
+        self,
+        embeddings: list[list[float]],
+        metadatas: Optional[list[dict[Any, Any]]] = None,
+        index_params: Optional[dict] = None,
+    ):
+        lock_name = f"vector_indexing_lock_{self._collection_name}"
+        with redis_client.lock(lock_name, timeout=20):
+            collection_exist_cache_key = f"vector_indexing_{self._collection_name}"
+            if redis_client.get(collection_exist_cache_key):
+                logger.info(f"Collection {self._collection_name} already exists.")
+                return
+
+            if not self._client.indices.exists(index=self._collection_name):
+                dim = len(embeddings[0])
+                settings = {
+                    "analysis": {
+                        "analyzer": {
+                            "ja_analyzer": {
+                                "type": "custom",
+                                "char_filter": [
+                                    "icu_normalizer",
+                                    "kuromoji_iteration_mark",
+                                ],
+                                "tokenizer": "kuromoji_tokenizer",
+                                "filter": [
+                                    "kuromoji_baseform",
+                                    "kuromoji_part_of_speech",
+                                    "ja_stop",
+                                    "kuromoji_number",
+                                    "kuromoji_stemmer",
+                                ],
+                            }
+                        }
+                    }
+                }
+                mappings = {
+                    "properties": {
+                        Field.CONTENT_KEY.value: {
+                            "type": "text",
+                            "analyzer": "ja_analyzer",
+                            "search_analyzer": "ja_analyzer",
+                        },
+                        Field.VECTOR.value: {  # Make sure the dimension is correct here
+                            "type": "dense_vector",
+                            "dims": dim,
+                            "index": True,
+                            "similarity": "cosine",
+                        },
+                        Field.METADATA_KEY.value: {
+                            "type": "object",
+                            "properties": {
+                                "doc_id": {"type": "keyword"}  # Map doc_id to keyword type
+                            },
+                        },
+                    }
+                }
+                self._client.indices.create(index=self._collection_name, settings=settings, mappings=mappings)
+
+            redis_client.set(collection_exist_cache_key, 1, ex=3600)
+
+
+class ElasticSearchJaVectorFactory(ElasticSearchVectorFactory):
+    def init_vector(self, dataset: Dataset, attributes: list, embeddings: Embeddings) -> ElasticSearchJaVector:
+        if dataset.index_struct_dict:
+            class_prefix: str = dataset.index_struct_dict["vector_store"]["class_prefix"]
+            collection_name = class_prefix
+        else:
+            dataset_id = dataset.id
+            collection_name = Dataset.gen_collection_name_by_id(dataset_id)
+            dataset.index_struct = json.dumps(self.gen_index_struct_dict(VectorType.ELASTICSEARCH, collection_name))
+
+        config = current_app.config
+        return ElasticSearchJaVector(
+            index_name=collection_name,
+            config=ElasticSearchConfig(
+                host=config.get("ELASTICSEARCH_HOST", "localhost"),
+                port=config.get("ELASTICSEARCH_PORT", 9200),
+                username=config.get("ELASTICSEARCH_USERNAME", ""),
+                password=config.get("ELASTICSEARCH_PASSWORD", ""),
+            ),
+            attributes=[],
+        )

api/core/rag/datasource/vdb/field.py

@@ -6,6 +6,8 @@ class Field(Enum):
     METADATA_KEY = "metadata"
     GROUP_KEY = "group_id"
     VECTOR = "vector"
+    # Sparse Vector aims to support full text search
+    SPARSE_VECTOR = "sparse_vector"
     TEXT_KEY = "text"
     PRIMARY_KEY = "id"
     DOC_ID = "metadata.doc_id"