Skip to content

knowledge_embedder.md

Latest commit

 

History

History
96 lines (68 loc) · 2.76 KB

knowledge_embedder.md

File metadata and controls

96 lines (68 loc) · 2.76 KB

Embedder

Embedders map unstructured data such as text and images into high-dimensional vector representations, enabling semantic similarity computation and comparison. They are a core component of semantic search in knowledge retrieval systems.

Below is an introduction to some commonly used components:

For more components, refer to Langchain Embedding models.

HuggingFaceEmbeddings

Install Dependencies

pip install --upgrade --quiet langchain langchain-huggingface sentence_transformers

Usage

  1. Create a HuggingFaceEmbeddings object
from langchain_huggingface import HuggingFaceEmbeddings

# Specify the HuggingFace model name to use
model_name = "BAAI/bge-small-en-v1.5"
# Specify model loading parameters; here it is set to run on CPU
model_kwargs = {"device": "cpu"}
# Specify encoding parameters; here it is set to normalize the output embeddings
encode_kwargs = {"normalize_embeddings": True}
# Create the HuggingFaceEmbeddings embedder object
embedder = HuggingFaceEmbeddings(
    model_name=model_name,
    model_kwargs=model_kwargs,
    encode_kwargs=encode_kwargs
)
  1. Construct a LangchainKnowledge object using this embedder object
from trpc_agent_sdk.server.knowledge.langchain_knowledge import LangchainKnowledge

rag = LangchainKnowledge(
    prompt_template=rag_prompt,
    document_loader=text_loader,
    document_transformer=text_splitter,
    embedder=embedder,  # Pass the constructed embedder
    vectorstore=vectorstore,
)

Reference

HunyuanEmbeddings

Install Dependencies

pip install hunyuan langchain-community
pip install "tencentcloud-sdk-python>=3.0.1139"

Usage

  1. Create a HunyuanEmbeddings object
from langchain_community.embeddings import HunyuanEmbeddings

embedder = HunyuanEmbeddings(
    hunyuan_secret_id="xxx",   # Hunyuan Secret ID, or set via the HUNYUAN_SECRET_ID environment variable
    hunyuan_secret_key="xxx",  # Hunyuan Secret Key, or set via the HUNYUAN_SECRET_KEY environment variable
    region="ap-guangzhou"      # Region of the Hunyuan service
)
  1. Construct a LangchainKnowledge object using this embedder object
rag = LangchainKnowledge(
    ...,
    embedder=embedder,
    ...,
)

Reference