Neural Notes: ColBERT & ColBERTv2

Created at 11am, Jan 16

benjamin

Artificial Intelligence

Contract ID

myQsZdr2jKqfKMDBdoDr7-l_SOrOen43qrBbXNcBN44

File Type

MP3

Entry Count

296

Embed. Model

jina_embeddings_v2_base_en

Index Type

hnsw

A podcast by Vertex Ventures US. A deep dive conversation on seminal research papers about neural information retrieval with one of the paper's authors and Ph.D. student at Stanford, Omar Khattab.ColBERT: Efficient and Effective Passage Search via Contextualized Late Interaction over BERTColBERTv2: Effective and Efficient Retrieval via Lightweight Late InteractionOriginal video: https://www.youtube.com/watch?v=8e3x5D_F-7c

And what Colbert says is actually you get to have both.

start: 15:42 - end: 15:44

So we are able to have highly scalable search because we've decomposed the problem into these summation of maximums.

And these are very easy to scale.

start: 15:52 - end: 15:55

But we're actually still scoring things contextually at the level of tokens that are able to interact with one another.

How to Retrieve?

# Search

curl -X POST "https://search.dria.co/hnsw/search" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"rerank": true, "top_n": 10, "contract_id": "myQsZdr2jKqfKMDBdoDr7-l_SOrOen43qrBbXNcBN44", "query": "What is alexanDRIA library?"}'
        
# Query

curl -X POST "https://search.dria.co/hnsw/query" \
-H "x-api-key: <YOUR_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"vector": [0.123, 0.5236], "top_n": 10, "contract_id": "myQsZdr2jKqfKMDBdoDr7-l_SOrOen43qrBbXNcBN44", "level": 2}'