feat: add sqlalchemy vectorizer helper #265
Conversation
Askir
force-pushed
the
jascha/add-vectorizer-field
branch
from
9216152 to
517ce9f
Compare
Askir
force-pushed
the
jascha/add-vectorizer-field
branch
from
8742af8 to
36cf4d9
Compare
Askir
changed the title
docs/python-integration.md
Outdated
| return False | ||
| return True | ||
|
|
||
| context.configure( |
There was a problem hiding this comment.
where is context coming from here?
There was a problem hiding this comment.
context is the alembic context used to configure and run migrations programmatically. Alembic autogenerates a fitting env.py for the user. E.g. this is how it looks if you're using async sqlalchemy: https://github.com/sqlalchemy/alembic/blob/main/alembic/templates/async/env.py
I can add the from alembic import context if you'd like?
| self._embedding_class = self.create_embedding_class(objtype) | ||
|
|
||
| # Set up relationship if requested | ||
| if self.add_relationship: |
There was a problem hiding this comment.
Instead of configuring whether or not to add a relationship, why don't we always add it but allow the user to define the lazy behavior and default that to select? With select nothing is ever pre-loaded right?
There was a problem hiding this comment.
Other suggestion, what if we simply forward all **kwargs, *args to the relationship function?
SQLAlchemy relationships are extremely configurable so I think just exposing anything the user might need while transparently adding the embedding model might be the cleanest interface.
This would basically make the Vectorizer helper an embedding_relationship function? We could even rename it as such?
There was a problem hiding this comment.
If you add the relationship, does it mean that related data is prefetched with a join?
There was a problem hiding this comment.
that's exactly what we are discussing right now. If you set lazy="select" it is only fetched as soon as it is accessed. if lazy is set to "joined" it's eagerly joined. There is also "selectin" for a separate query with an in statement, and a few other options.
Askir
force-pushed
the
jascha/add-vectorizer-field
branch
3 times, most recently
from
e342f0b to
c108a87
Compare
| ); | ||
| ``` | ||
|
|
||
| We recommend adding this to a migration script and run it via alembic. |
There was a problem hiding this comment.
Let's add another reference to the Working with alembic section below.
docs/python-integration.md
Outdated
| - `dimensions` (int): The size of the embedding vector (required) | ||
| - `target_schema` (str, optional): Override the schema for the embeddings table. If not provided, inherits from the parent model's schema | ||
| - `target_table` (str, optional): Override the table name for embeddings. Default is `{table_name}_embedding_store` | ||
| - `add_relationship` (bool, optional): Whether to automatically create a relationship to the embeddings table (default: False) |
There was a problem hiding this comment.
Let's add a reference to the ### 2. Relationship Access section below. At this point I didn't know what are sqlalchemy relationships, and if I needed/wanted one or not.
Askir
force-pushed
the
jascha/add-vectorizer-field
branch
7 times, most recently
from
3b47afc to
8fe145e
Compare
docs/python-integration.md
Outdated
|
|
||
| Additional parameters are simply forwarded to the underlying [SQLAlchemy relationship](https://docs.sqlalchemy.org/en/20/orm/relationships.html) so you can configure it as you desire. | ||
|
|
||
| Think of the `embedding_relationship` as a normal SQLAlchemy relationship, but with a preconfigured model instance under the hood. |
There was a problem hiding this comment.
Also maybe something similar belongs in the intro?
| Think of the `embedding_relationship` as a normal SQLAlchemy relationship, but with a preconfigured model instance under the hood. | |
| Think of the `embedding_relationship` as a normal SQLAlchemy relationship, but with a preconfigured embedding model instance under the hood. |
Askir
force-pushed
the
jascha/add-vectorizer-field
branch
from
8fe145e to
882f91e
Compare
Askir
force-pushed
the
jascha/add-vectorizer-field
branch
from
4dae567 to
22233a0
Compare
This PR adds a helper descriptorVectorizerthat creates an SQLAlchemy Model which can be used to reference vectorizer managed embedding tables without explicitly configuring it yourself.I've done another pass at the interface and now have renamed the
Vectorizertoembedding_relationshipdetailed in this discussion:#265 (comment)