mirror of
https://github.com/danswer-ai/danswer.git
synced 2025-06-15 02:20:52 +02:00
39 lines
1.2 KiB
Python
39 lines
1.2 KiB
Python
import math
|
|
import uuid
|
|
|
|
from danswer.chunking.models import IndexChunk
|
|
from danswer.chunking.models import InferenceChunk
|
|
|
|
|
|
DEFAULT_BATCH_SIZE = 30
|
|
|
|
|
|
def translate_boost_count_to_multiplier(boost: int) -> float:
|
|
"""Mapping boost integer values to a multiplier according to a sigmoid curve
|
|
Piecewise such that at many downvotes, its 0.5x the score and with many upvotes
|
|
it is 2x the score. This should be in line with the Vespa calculation."""
|
|
# 3 in the equation below stretches it out to hit asymptotes slower
|
|
if boost < 0:
|
|
# 0.5 + sigmoid -> range of 0.5 to 1
|
|
return 0.5 + (1 / (1 + math.exp(-1 * boost / 3)))
|
|
|
|
# 2 x sigmoid -> range of 1 to 2
|
|
return 2 / (1 + math.exp(-1 * boost / 3))
|
|
|
|
|
|
def get_uuid_from_chunk(
|
|
chunk: IndexChunk | InferenceChunk, mini_chunk_ind: int = 0
|
|
) -> uuid.UUID:
|
|
doc_str = (
|
|
chunk.document_id
|
|
if isinstance(chunk, InferenceChunk)
|
|
else chunk.source_document.id
|
|
)
|
|
# Web parsing URL duplicate catching
|
|
if doc_str and doc_str[-1] == "/":
|
|
doc_str = doc_str[:-1]
|
|
unique_identifier_string = "_".join(
|
|
[doc_str, str(chunk.chunk_id), str(mini_chunk_ind)]
|
|
)
|
|
return uuid.uuid5(uuid.NAMESPACE_X500, unique_identifier_string)
|