Spaces:
Runtime error
Runtime error
File size: 2,574 Bytes
129cd69 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 |
"""KNN Retriever.
Largely based on
https://github.com/karpathy/randomfun/blob/master/knn_vs_svm.ipynb"""
from __future__ import annotations
import concurrent.futures
from typing import Any, List, Optional
import numpy as np
from langchain_core.documents import Document
from langchain_core.embeddings import Embeddings
from langchain_core.retrievers import BaseRetriever
from langchain.callbacks.manager import CallbackManagerForRetrieverRun
def create_index(contexts: List[str], embeddings: Embeddings) -> np.ndarray:
"""
Create an index of embeddings for a list of contexts.
Args:
contexts: List of contexts to embed.
embeddings: Embeddings model to use.
Returns:
Index of embeddings.
"""
with concurrent.futures.ThreadPoolExecutor() as executor:
return np.array(list(executor.map(embeddings.embed_query, contexts)))
class KNNRetriever(BaseRetriever):
"""`KNN` retriever."""
embeddings: Embeddings
"""Embeddings model to use."""
index: Any
"""Index of embeddings."""
texts: List[str]
"""List of texts to index."""
k: int = 4
"""Number of results to return."""
relevancy_threshold: Optional[float] = None
"""Threshold for relevancy."""
class Config:
"""Configuration for this pydantic object."""
arbitrary_types_allowed = True
@classmethod
def from_texts(
cls, texts: List[str], embeddings: Embeddings, **kwargs: Any
) -> KNNRetriever:
index = create_index(texts, embeddings)
return cls(embeddings=embeddings, index=index, texts=texts, **kwargs)
def _get_relevant_documents(
self, query: str, *, run_manager: CallbackManagerForRetrieverRun
) -> List[Document]:
query_embeds = np.array(self.embeddings.embed_query(query))
# calc L2 norm
index_embeds = self.index / np.sqrt((self.index**2).sum(1, keepdims=True))
query_embeds = query_embeds / np.sqrt((query_embeds**2).sum())
similarities = index_embeds.dot(query_embeds)
sorted_ix = np.argsort(-similarities)
denominator = np.max(similarities) - np.min(similarities) + 1e-6
normalized_similarities = (similarities - np.min(similarities)) / denominator
top_k_results = [
Document(page_content=self.texts[row])
for row in sorted_ix[0 : self.k]
if (
self.relevancy_threshold is None
or normalized_similarities[row] >= self.relevancy_threshold
)
]
return top_k_results
|