## Description
This PR adds support for Memcached as a usable LLM model cache by adding
the ```MemcachedCache``` implementation relying on the
[pymemcache](https://github.com/pinterest/pymemcache) client.
Unit test-wise, the new integration is generally covered under existing
import testing. All new functionality depends on pymemcache if
instantiated and used, so to comply with the other cache implementations
the PR also adds optional integration tests for ```MemcachedCache```.
Since this is a new integration, documentation is added for Memcached as
an integration and as an LLM Cache.
## Issue
This PR closes#27275 which was originally raised as a discussion in
#27035
## Dependencies
There are no new required dependencies for langchain, but
[pymemcache](https://github.com/pinterest/pymemcache) is required to
instantiate the new ```MemcachedCache```.
## Example Usage
```python3
from langchain.globals import set_llm_cache
from langchain_openai import OpenAI
from langchain_community.cache import MemcachedCache
from pymemcache.client.base import Client
llm = OpenAI(model="gpt-3.5-turbo-instruct", n=2, best_of=2)
set_llm_cache(MemcachedCache(Client('localhost')))
# The first time, it is not yet in cache, so it should take longer
llm.invoke("Which city is the most crowded city in the USA?")
# The second time it is, so it goes faster
llm.invoke("Which city is the most crowded city in the USA?")
```
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
This PR adds a `SingleStoreDBSemanticCache` class that implements a
cache based on SingleStoreDB vector store, integration tests, and a
notebook example.
Additionally, this PR contains minor changes to SingleStoreDB vector
store:
- change add texts/documents methods to return a list of inserted ids
- implement delete(ids) method to delete documents by list of ids
- added drop() method to drop a correspondent database table
- updated integration tests to use and check functionality implemented
above
CC: @baskaryan, @hwchase17
---------
Co-authored-by: Volodymyr Tkachuk <vtkachuk-ua@singlestore.com>
This PR improves on the `CassandraCache` and `CassandraSemanticCache`
classes, mainly in the constructor signature, and also introduces
several minor improvements around these classes.
### Init signature
A (sigh) breaking change is tentatively introduced to the constructor.
To me, the advantages outweigh the possible discomfort: the new syntax
places the DB-connection objects `session` and `keyspace` later in the
param list, so that they can be given a default value. This is what
enables the pattern of _not_ specifying them, provided one has
previously initialized the Cassandra connection through the versatile
utility method `cassio.init(...)`.
In this way, a much less unwieldy instantiation can be done, such as
`CassandraCache()` and `CassandraSemanticCache(embedding=xyz)`,
everything else falling back to defaults.
A downside is that, compared to the earlier signature, this might turn
out to be breaking for those doing positional instantiation. As a way to
mitigate this problem, this PR typechecks its first argument trying to
detect the legacy usage.
(And to make this point less tricky in the future, most arguments are
left to be keyword-only).
If this is considered too harsh, I'd like guidance on how to further
smoothen this transition. **Our plan is to make the pattern of optional
session/keyspace a standard across all Cassandra classes**, so that a
repeatable strategy would be ideal. A possibility would be to keep
positional arguments for legacy reasons but issue a deprecation warning
if any of them is actually used, to later remove them with 0.2 - please
advise on this point.
### Other changes
- class docstrings: enriched, completely moved to class level, added
note on `cassio.init(...)` pattern, added tiny sample usage code.
- semantic cache: revised terminology to never mention "distance" (it is
in fact a similarity!). Kept the legacy constructor param with a
deprecation warning if used.
- `llm_caching` notebook: uniform flow with the Cassandra and Astra DB
separate cases; better and Cassandra-first description; all imports made
explicit and from community where appropriate.
- cache integration tests moved to community (incl. the imported tools),
env var bugfix for `CASSANDRA_CONTACT_POINTS`.
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
0.2rc
migrations
- [x] Move memory
- [x] Move remaining retrievers
- [x] graph_qa chains
- [x] some dependency from evaluation code potentially on math utils
- [x] Move openapi chain from `langchain.chains.api.openapi` to
`langchain_community.chains.openapi`
- [x] Migrate `langchain.chains.ernie_functions` to
`langchain_community.chains.ernie_functions`
- [x] migrate `langchain/chains/llm_requests.py` to
`langchain_community.chains.llm_requests`
- [x] Moving `langchain_community.cross_enoders.base:BaseCrossEncoder`
->
`langchain_community.retrievers.document_compressors.cross_encoder:BaseCrossEncoder`
(namespace not ideal, but it needs to be moved to `langchain` to avoid
circular deps)
- [x] unit tests langchain -- add pytest.mark.community to some unit
tests that will stay in langchain
- [x] unit tests community -- move unit tests that depend on community
to community
- [x] mv integration tests that depend on community to community
- [x] mypy checks
Other todo
- [x] Make deprecation warnings not noisy (need to use warn deprecated
and check that things are implemented properly)
- [x] Update deprecation messages with timeline for code removal (likely
we actually won't be removing things until 0.4 release) -- will give
people more time to transition their code.
- [ ] Add information to deprecation warning to show users how to
migrate their code base using langchain-cli
- [ ] Remove any unnecessary requirements in langchain (e.g., is
SQLALchemy required?)
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
