feat(mode) : Adding the siliconflow embeding proxy (#2603)

Co-authored-by: 9527 <9527@163.com>
Co-authored-by: Fangyin Cheng <staneyffer@gmail.com>

configs/dbgpt-proxy-siliconflow-mysql.toml

@@ -33,9 +33,8 @@ provider = "proxy/siliconflow"
 api_key = "${env:SILICONFLOW_API_KEY}"
 
 [[models.embeddings]]
-name = "BAAI/bge-large-zh-v1.5"
-provider = "proxy/openai"
-api_url = "https://api.siliconflow.cn/v1/embeddings"
+name = "BAAI/bge-m3"
+provider = "proxy/siliconflow"
 api_key = "${env:SILICONFLOW_API_KEY}"
 
 [[models.rerankers]]

packages/dbgpt-ext/src/dbgpt_ext/rag/embeddings/__init__.py

@@ -3,6 +3,7 @@
 from .jina import JinaEmbeddings  # noqa: F401
 from .ollama import OllamaEmbeddings  # noqa: F401
 from .qianfan import QianFanEmbeddings  # noqa: F401
+from .siliconflow import SiliconFlowEmbeddings  # noqa: F401
 from .tongyi import TongYiEmbeddings  # noqa: F401
 
 __ALL__ = [
@@ -10,4 +11,5 @@ __ALL__ = [
     "OllamaEmbeddings",
     "QianFanEmbeddings",
     "TongYiEmbeddings",
+    "SiliconFlowEmbeddings",
 ]

packages/dbgpt-ext/src/dbgpt_ext/rag/embeddings/siliconflow.py

@@ -0,0 +1,150 @@
+from dataclasses import dataclass, field
+from typing import List, Optional, Type
+
+from dbgpt._private.pydantic import BaseModel, ConfigDict, Field
+from dbgpt.core import EmbeddingModelMetadata, Embeddings
+from dbgpt.core.interface.parameter import EmbeddingDeployModelParameters
+from dbgpt.model.adapter.base import register_embedding_adapter
+from dbgpt.util.i18n_utils import _
+
+
+@dataclass
+class SiliconFlowEmbeddingDeployModelParameters(EmbeddingDeployModelParameters):
+    """SiliconFlow Embeddings deploy model parameters."""
+
+    provider: str = "proxy/siliconflow"
+
+    api_key: Optional[str] = field(
+        default="${env:SILICONFLOW_API_KEY}",
+        metadata={
+            "help": _("The API key for the rerank API."),
+        },
+    )
+    backend: Optional[str] = field(
+        default="BAAI/bge-m3",
+        metadata={
+            "help": _(
+                "The real model name to pass to the provider, default is None. If "
+                "backend is None, use name as the real model name."
+            ),
+        },
+    )
+
+    @property
+    def real_provider_model_name(self) -> str:
+        """Get the real provider model name."""
+        return self.backend or self.name
+
+
+class SiliconFlowEmbeddings(BaseModel, Embeddings):
+    """The SiliconFlow embeddings."""
+
+    model_config = ConfigDict(arbitrary_types_allowed=True, protected_namespaces=())
+    api_key: Optional[str] = Field(
+        default=None, description="The API key for the embeddings API."
+    )
+    model_name: str = Field(
+        default="BAAI/bge-m3", description="The name of the model to use."
+    )
+
+    def __init__(self, **kwargs):
+        """Initialize the SiliconFlow Embeddings."""
+        super().__init__(**kwargs)
+        self._api_key = self.api_key
+
+    @classmethod
+    def param_class(cls) -> Type[SiliconFlowEmbeddingDeployModelParameters]:
+        """Get the parameter class."""
+        return SiliconFlowEmbeddingDeployModelParameters
+
+    @classmethod
+    def from_parameters(
+        cls, parameters: SiliconFlowEmbeddingDeployModelParameters
+    ) -> "Embeddings":
+        """Create an embedding model from parameters."""
+        return cls(
+            api_key=parameters.api_key,
+            model_name=parameters.real_provider_model_name,
+        )
+
+    def embed_documents(
+        self, texts: List[str], max_batch_chunks_size=25
+    ) -> List[List[float]]:
+        """Get the embeddings for a list of texts.
+
+        Args:
+            texts (List[str]): A list of texts to get embeddings for.
+            max_batch_chunks_size: The max batch size for embedding.
+
+        Returns:
+            Embedded texts as List[List[float]], where each inner List[float]
+                corresponds to a single input text.
+        """
+        import requests
+
+        embeddings = []
+        headers = {"Authorization": f"Bearer {self._api_key}"}
+
+        for i in range(0, len(texts), max_batch_chunks_size):
+            batch_texts = texts[i : i + max_batch_chunks_size]
+            response = requests.post(
+                url="https://api.siliconflow.cn/v1/embeddings",
+                json={"model": self.model_name, "input": batch_texts},
+                headers=headers,
+            )
+
+            if response.status_code != 200:
+                raise RuntimeError(f"Embedding failed: {response.text}")
+
+            # 提取并排序嵌入
+            data = response.json()
+            batch_embeddings = data["data"]
+            sorted_embeddings = sorted(batch_embeddings, key=lambda e: e["index"])
+            embeddings.extend([result["embedding"] for result in sorted_embeddings])
+
+        return embeddings
+
+    def embed_query(self, text: str) -> List[float]:
+        """Compute query embeddings using a SiliconFlow embedding model.
+
+        Args:
+            text: The text to embed.
+
+        Returns:
+            Embeddings for the text.
+        """
+        return self.embed_documents([text])[0]
+
+
+register_embedding_adapter(
+    SiliconFlowEmbeddings,
+    supported_models=[
+        EmbeddingModelMetadata(
+            model="BAAI/bge-m3",
+            dimension=1024,
+            context_length=8192,
+            description=_(
+                "The embedding model is provided by SiliconFlow, supporting multiple "
+                "languages and high-quality text embeddings."
+            ),
+        ),
+        EmbeddingModelMetadata(
+            model="BAAI/bge-large-zh-v1.5",
+            dimension=1024,
+            context_length=512,
+            description=_(
+                "The embedding model is provided by SiliconFlow, supporting multiple "
+                "languages and high-quality text embeddings."
+            ),
+        ),
+        EmbeddingModelMetadata(
+            model="BAAI/bge-large-en-v1.5",
+            dimension=1024,
+            context_length=512,
+            description=_(
+                "The embedding model is provided by SiliconFlow, supporting multiple "
+                "languages and high-quality text embeddings."
+            ),
+        ),
+    ],
+)