Sam
b2b01185f2
* FEATURE: add support for new OpenAI embedding models This adds support for just released text_embedding_3_small and large Note, we have not yet implemented truncation support which is a new API feature. (triggered using dimensions) * Tiny side fix, recalc bots when ai is enabled or disabled * FIX: downsample to 2000 items per vector which is a pgvector limitation
54 lines
1.0 KiB
Ruby
54 lines
1.0 KiB
Ruby
# frozen_string_literal: true
|
|
|
|
module DiscourseAi
|
|
module Embeddings
|
|
module VectorRepresentations
|
|
class TextEmbedding3Large < Base
|
|
def id
|
|
7
|
|
end
|
|
|
|
def version
|
|
1
|
|
end
|
|
|
|
def name
|
|
"text-embedding-3-large"
|
|
end
|
|
|
|
def dimensions
|
|
# real dimentions are 3072, but we only support up to 2000 in the
|
|
# indexes, so we downsample to 2000 via API
|
|
2000
|
|
end
|
|
|
|
def max_sequence_length
|
|
8191
|
|
end
|
|
|
|
def pg_function
|
|
"<=>"
|
|
end
|
|
|
|
def pg_index_type
|
|
"vector_cosine_ops"
|
|
end
|
|
|
|
def vector_from(text)
|
|
response =
|
|
DiscourseAi::Inference::OpenAiEmbeddings.perform!(
|
|
text,
|
|
model: name,
|
|
dimensions: dimensions,
|
|
)
|
|
response[:data].first[:embedding]
|
|
end
|
|
|
|
def tokenizer
|
|
DiscourseAi::Tokenizer::OpenAiTokenizer
|
|
end
|
|
end
|
|
end
|
|
end
|
|
end
|