Spaces:
Sleeping
Sleeping
Create utils.py
Browse filesSimple utils add red on iPad
utils.py
ADDED
|
@@ -0,0 +1,42 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import chromadb
|
| 2 |
+
from sentence_transformers import CrossEncoder, SentenceTransformer
|
| 3 |
+
|
| 4 |
+
def chroma_client_setup():
|
| 5 |
+
chroma_client = chromadb.Client()
|
| 6 |
+
collection = client.create_collection(
|
| 7 |
+
name="food_collection",
|
| 8 |
+
metadata={"hnsw:space": "cosine"} # l2 is the default
|
| 9 |
+
)
|
| 10 |
+
return collection
|
| 11 |
+
|
| 12 |
+
def embedding_function(items_to_embed: list[str]):
|
| 13 |
+
sentence_model = SentenceTransformer(
|
| 14 |
+
"mixedbread-ai/mxbai-embed-large-v1"
|
| 15 |
+
)
|
| 16 |
+
embedded_items = sentence_model.encode(
|
| 17 |
+
items_to_embed,
|
| 18 |
+
show_progress_bar=True
|
| 19 |
+
)
|
| 20 |
+
return embedded_items
|
| 21 |
+
|
| 22 |
+
def chroma_upserting(collection, embeddings:list[list[str]], payload:list[dict]):
|
| 23 |
+
collection.add(
|
| 24 |
+
documents=[item['doc'] for item in payload],
|
| 25 |
+
embeddings=embeddings,
|
| 26 |
+
metadatas=payload,
|
| 27 |
+
ids=[f"id{item}" for item in range(len(embedfings))]
|
| 28 |
+
)
|
| 29 |
+
|
| 30 |
+
def search_chroma(collection, query:str):
|
| 31 |
+
results = collection.query(
|
| 32 |
+
query_embeddings=embedding_function([query]),
|
| 33 |
+
n_results=5
|
| 34 |
+
)
|
| 35 |
+
return results
|
| 36 |
+
|
| 37 |
+
def reranking_results(query: str, top_k_results: list[str]):
|
| 38 |
+
# Load the model, here we use our base sized model
|
| 39 |
+
rerank_model = CrossEncoder("mixedbread-ai/mxbai-rerank-xsmall-v1")
|
| 40 |
+
reranked_results = rerank_model.rank(query, top_k_results, return_documents=True)
|
| 41 |
+
return reranked_results
|
| 42 |
+
|