feat: support multivector type #2005
Conversation
Signed-off-by: BubbleCal <bubble-cal@outlook.com>
…ch 'main' of https://github.com/lancedb/lancedb into multivec Signed-off-by: BubbleCal <bubble-cal@outlook.com>
Signed-off-by: BubbleCal <bubble-cal@outlook.com>
| })?; | ||
|
|
||
| let mut is_binary = false; | ||
| if let arrow_schema::DataType::FixedSizeList(element, dim) = field.data_type() { |
There was a problem hiding this comment.
lance would check this, so remove it
Signed-off-by: BubbleCal <bubble-cal@outlook.com>
Signed-off-by: BubbleCal <bubble-cal@outlook.com>
python/python/tests/test_query.py
Outdated
| def multivec_table(tmp_path) -> lancedb.table.Table: | ||
| db = lancedb.connect(tmp_path) |
There was a problem hiding this comment.
Let's not create this on disk if we don't need it to be there
| def multivec_table(tmp_path) -> lancedb.table.Table: | |
| db = lancedb.connect(tmp_path) | |
| def multivec_table() -> lancedb.table.Table: | |
| db = lancedb.connect("memory://") |
python/python/tests/test_query.py
Outdated
| async def multivec_table_async(tmp_path) -> AsyncTable: | ||
| conn = await lancedb.connect_async( | ||
| tmp_path, read_consistency_interval=timedelta(seconds=0) | ||
| ) |
There was a problem hiding this comment.
| async def multivec_table_async(tmp_path) -> AsyncTable: | |
| conn = await lancedb.connect_async( | |
| tmp_path, read_consistency_interval=timedelta(seconds=0) | |
| ) | |
| async def multivec_table_async() -> AsyncTable: | |
| conn = await lancedb.connect_async( | |
| "memory://", read_consistency_interval=timedelta(seconds=0) | |
| ) |
docs/src/search.md
Outdated
|
|
||
| You can index on a column with multivector type and search on it, the query can be single vector or multiple vectors. If the query is multiple vectors `mq`, the similarity (distance) from it to any multivector `mv` in the dataset, is defined as: | ||
|
|
||
| **maxsim(mq, mv) = sum(max(sim(mq[i], mv[j])))** |
There was a problem hiding this comment.
It's unclear what the max and sum are over. I think you are missing that part of the formula, right?
There was a problem hiding this comment.
it seems a math formula is not supported, I will post an image here
There was a problem hiding this comment.
You could add the math plugin if you want: https://squidfunk.github.io/mkdocs-material/reference/math/#katex (I'd prefer the katex over mathjax, I think)
Signed-off-by: BubbleCal <bubble-cal@outlook.com>
|
|
||
| LanceDB supports multivector type, this is useful when you have multiple vectors for a single item (e.g. with ColBert and ColPali). | ||
|
|
||
| You can index on a column with multivector type and search on it, the query can be single vector or multiple vectors. If the query is multiple vectors `mq`, the similarity (distance) from it to any multivector `mv` in the dataset, is defined as: |
There was a problem hiding this comment.
Should we say similarity is 1 - distance?
| @@ -138,6 +138,36 @@ LanceDB supports binary vectors as a data type, and has the ability to search bi | |||
| --8<-- "python/python/tests/docs/test_binary_vector.py:async_binary_vector" | |||
| ``` | |||
|
|
|||
| ## Multivector type | |||
|
|
|||
| LanceDB supports multivector type, this is useful when you have multiple vectors for a single item (e.g. with ColBert and ColPali). | |||
There was a problem hiding this comment.
This is still missing an intuitive explanation of what this query type means, and how it's different from the single vector situation. Could you perhaps root this in an example? Is the main example chunked documents, where each row is a full document, and each vector made for a chunk?
| === "sync API" | ||
|
|
||
| ```python | ||
| --8<-- "python/python/tests/docs/test_multivector.py:imports" | ||
|
|
||
| --8<-- "python/python/tests/docs/test_multivector.py:sync_multivector" | ||
| ``` | ||
|
|
||
| === "async API" | ||
|
|
||
| ```python | ||
| --8<-- "python/python/tests/docs/test_multivector.py:imports" | ||
|
|
||
| --8<-- "python/python/tests/docs/test_multivector.py:async_multivector" | ||
| ``` |
There was a problem hiding this comment.
I just realized these files are missing. Do you have a copy of them somewhere @BubbleCal ?
No description provided.