mirror of
https://github.com/trustgraph-ai/trustgraph.git
synced 2026-04-25 16:36:21 +02:00
a3ea1301d6
- Break out store queries, so not locked into a Milvus/Cassandra backend - Break out prompting into a separate module, so that prompts can be tailored to other LLMs - Jsonnet used to generate docker compose templates - Version to 0.6.0
35 lines
903 B
Python
Executable file
35 lines
903 B
Python
Executable file
#!/usr/bin/env python3
|
|
|
|
from langchain_huggingface import HuggingFaceEmbeddings
|
|
|
|
from trustgraph.direct.milvus import TripleVectors
|
|
|
|
client = TripleVectors()
|
|
|
|
embeddings = HuggingFaceEmbeddings(model_name="all-MiniLM-L6-v2")
|
|
|
|
text="""A cat is a small animal. A dog is a large animal.
|
|
Cats say miaow. Dogs go woof.
|
|
"""
|
|
|
|
embeds = embeddings.embed_documents([text])[0]
|
|
|
|
text2="""If you couldn't download the model due to network issues, as a walkaround, you can use random vectors to represent the text and still finish the example. Just note that the search result won't reflect semantic similarity as the vectors are fake ones.
|
|
"""
|
|
|
|
embeds2 = embeddings.embed_documents([text2])[0]
|
|
|
|
client.insert(embeds, "animals")
|
|
client.insert(embeds, "vectors")
|
|
|
|
query="""What noise does a cat make?"""
|
|
|
|
qembeds = embeddings.embed_documents([query])[0]
|
|
|
|
res = client.search(
|
|
qembeds,
|
|
limit=2
|
|
)
|
|
|
|
print(res)
|
|
|