Selecting embedding size in BedrockEmbedding titan v1 and v2, switching dimensions dynamically via keyword arguments using langchain

I am trying to provide dimension argument while trying to fetch the Titan version 1 and version 2 embedding models, but the BedrockEmbedding class does not have the feature of dimension, It only provideds the default 1024 dimension. How can we solve this?

Below is how i am calling BedrockEmbedding from langchain_community.embeddings import BedrockEmbeddings

トピック: 機械学習と AI AWSでのジェネレーティブAI
タグ: ラングチェイン Amazon Bedrock
言語: English

Manish

質問済み 2年前878ビュー

1回答

新しい順
投票が多い順
コメントが多い順

Hi,

You don't have to provide the embeddings dimensions in the query: the model that you select (via its id) will return the number of dimensions that it is programmed for.

So, if you work with multiple embedding engines, your code must be ready to work with variable numbers of dimensions depending on the model that you query.

Best,

Didier

エキスパート

Didier Durand

回答済み 2年前

エキスパート

Oleksii Bebych

レビュー済み 2年前

エキスパート

kranthi putti

レビュー済み 2年前

Manish
2年前
Currently, our solution selects the model_id and defaults the text embedding dimensions to 1536 for Titan version 1 and 1024 for Titan version 2. However, we aim to test our dataset with various embedding sizes, such as 256, 512, and 1024.

How can we dynamically set the embedding dimension for both Titan version 1 and version 2 using the Langchain Community BedrockEmbedding? The model keyword arguments do not seem to accept dimension specifications.

How do we dynamically select the embedding dimension for both Titan version 1 and version 2 during testing.

Selecting embedding size in BedrockEmbedding titan v1 and v2, switching dimensions dynamically via keyword arguments using langchain

回答を追加する

関連するコンテンツ