<!--
Thank you for contributing to LangChain! Your PR will appear in our
release under the title you set. Please make sure it highlights your
valuable contribution.
Replace this with a description of the change, the issue it fixes (if
applicable), and relevant context. List any dependencies required for
this change.
After you're done, someone will review your PR. They may suggest
improvements. If no one reviews your PR within a few days, feel free to
@-mention the same people again, as notifications can get lost.
Finally, we'd love to show appreciation for your contribution - if you'd
like us to shout you out on Twitter, please also include your handle!
-->
#### Add start index to metadata in TextSplitter
- Modified method `create_documents` to track start position of each
chunk
- The `start_index` is included in the metadata if the `add_start_index`
parameter in the class constructor is set to `True`
This enables referencing back to the original document, particularly
useful when a specific chunk is retrieved.
<!-- If you're adding a new integration, please include:
1. a test for the integration - favor unit tests that does not rely on
network access.
2. an example notebook showing its use
See contribution guidelines for more information on how to write tests,
lint
etc:
https://github.com/hwchase17/langchain/blob/master/.github/CONTRIBUTING.md
-->
#### Who can review?
Tag maintainers/contributors who might be interested:
@eyurtsev @agola11
<!-- For a quicker response, figure out the right person to tag with @
@hwchase17 - project lead
Tracing / Callbacks
- @agola11
Async
- @agola11
DataLoaders
- @eyurtsev
Models
- @hwchase17
- @agola11
Agents / Tools / Toolkits
- @vowelparrot
VectorStores / Retrievers / Memory
- @dev2049
-->
This PR adds a Baseten integration. I've done my best to follow the
contributor's guidelines and add docs, an example notebook, and an
integration test modeled after similar integrations' test.
Please let me know if there is anything I can do to improve the PR. When
it is merged, please tag https://twitter.com/basetenco and
https://twitter.com/philip_kiely as contributors (the note on the PR
template said to include Twitter accounts)
<!--
Thank you for contributing to LangChain! Your PR will appear in our
release under the title you set. Please make sure it highlights your
valuable contribution.
Replace this with a description of the change, the issue it fixes (if
applicable), and relevant context. List any dependencies required for
this change.
After you're done, someone will review your PR. They may suggest
improvements. If no one reviews your PR within a few days, feel free to
@-mention the same people again, as notifications can get lost.
Finally, we'd love to show appreciation for your contribution - if you'd
like us to shout you out on Twitter, please also include your handle!
-->
<!-- Remove if not applicable -->
Fixes#3983
Mimicing what we do for saving and loading VectorDBQA chain, I added the
logic for RetrievalQA chain.
Also added a unit test. I did not find how we test other chains for
their saving and loading functionality, so I just added a file with one
test case. Let me know if there are recommended ways to test it.
#### Before submitting
<!-- If you're adding a new integration, please include:
1. a test for the integration - favor unit tests that does not rely on
network access.
2. an example notebook showing its use
See contribution guidelines for more information on how to write tests,
lint
etc:
https://github.com/hwchase17/langchain/blob/master/.github/CONTRIBUTING.md
-->
#### Who can review?
Tag maintainers/contributors who might be interested:
@dev2049
<!-- For a quicker response, figure out the right person to tag with @
@hwchase17 - project lead
Tracing / Callbacks
- @agola11
Async
- @agola11
DataLoaders
- @eyurtsev
Models
- @hwchase17
- @agola11
Agents / Tools / Toolkits
- @vowelparrot
VectorStores / Retrievers / Memory
- @dev2049
-->
---------
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
- Added `SingleStoreDB` vector store, which is a wrapper over the
SingleStore DB database, that can be used as a vector storage and has an
efficient similarity search.
- Added integration tests for the vector store
- Added jupyter notebook with the example
@dev2049
---------
Co-authored-by: Volodymyr Tkachuk <vtkachuk-ua@singlestore.com>
Co-authored-by: Dev 2049 <dev.dev2049@gmail.com>
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
Implementation of similarity_search_with_relevance_scores for quadrant
vector store.
As implemented the method is also compatible with other capacities such
as filtering.
Integration tests updated.
#### Who can review?
Tag maintainers/contributors who might be interested:
VectorStores / Retrievers / Memory
- @dev2049
### Summary
Adds an `UnstructuredCSVLoader` for loading CSVs. One advantage of using
`UnstructuredCSVLoader` relative to the standard `CSVLoader` is that if
you use `UnstructuredCSVLoader` in `"elements"` mode, an HTML
representation of the table will be available in the metadata.
#### Who can review?
@hwchase17
@eyurtsev
#### Who can review?
Tag maintainers/contributors who might be interested:
@hwchase17 - project lead
- @agola11
---------
Co-authored-by: Yessen Kanapin <yessen@deepinfra.com>
Fixes # 5807
Realigned tests with implementation.
Also reinforced folder unicity for the test_faiss_local_save_load test
using date-time suffix
#### Before submitting
- Integration test updated
- formatting and linting ok (locally)
#### Who can review?
Tag maintainers/contributors who might be interested:
@hwchase17 - project lead
VectorStores / Retrievers / Memory
-@dev2049
This introduces the `YoutubeAudioLoader`, which will load blobs from a
YouTube url and write them. Blobs are then parsed by
`OpenAIWhisperParser()`, as show in this
[PR](https://github.com/hwchase17/langchain/pull/5580), but we extend
the parser to split audio such that each chuck meets the 25MB OpenAI
size limit. As shown in the notebook, this enables a very simple UX:
```
# Transcribe the video to text
loader = GenericLoader(YoutubeAudioLoader([url],save_dir),OpenAIWhisperParser())
docs = loader.load()
```
Tested on full set of Karpathy lecture videos:
```
# Karpathy lecture videos
urls = ["https://youtu.be/VMj-3S1tku0"
"https://youtu.be/PaCmpygFfXo",
"https://youtu.be/TCH_1BHY58I",
"https://youtu.be/P6sfmUTpUmc",
"https://youtu.be/q8SA3rM6ckI",
"https://youtu.be/t3YJ5hKiMQ0",
"https://youtu.be/kCc8FmEb1nY"]
# Directory to save audio files
save_dir = "~/Downloads/YouTube"
# Transcribe the videos to text
loader = GenericLoader(YoutubeAudioLoader(urls,save_dir),OpenAIWhisperParser())
docs = loader.load()
```
# What does this PR do?
Change the HTML tags so that a tag with attributes can be found.
## Before submitting
- [x] Tests added
- [x] CI/CD validated
### Who can review?
Anyone in the community is free to review the PR once the tests have
passed. Feel free to tag
members/contributors who may be interested in your PR.
- Remove the client implementation (this breaks backwards compatibility
for existing testers. I could keep the stub in that file if we want, but
not many people are using it yet
- Add SDK as dependency
- Update the 'run_on_dataset' method to be a function that optionally
accepts a client as an argument
- Remove the langchain plus server implementation (you get it for free
with the SDK now)
We could make the SDK optional for now, but the plan is to use w/in the
tracer so it would likely become a hard dependency at some point.
Fixes#5614
#### Issue
The `***` combination produces an exception when used as a seperator in
`re.split`. Instead `\*\*\*` should be used for regex exprations.
#### Who can review?
@eyurtsev
Aviary is an open source toolkit for evaluating and deploying open
source LLMs. You can find out more about it on
[http://github.com/ray-project/aviary). You can try it out at
[http://aviary.anyscale.com](aviary.anyscale.com).
This code adds support for Aviary in LangChain. To minimize
dependencies, it connects directly to the HTTP endpoint.
The current implementation is not accelerated and uses the default
implementation of `predict` and `generate`.
It includes a test and a simple example.
@hwchase17 and @agola11 could you have a look at this?
---------
Co-authored-by: Dev 2049 <dev.dev2049@gmail.com>
# OpenAIWhisperParser
This PR creates a new parser, `OpenAIWhisperParser`, that uses the
[OpenAI Whisper
model](https://platform.openai.com/docs/guides/speech-to-text/quickstart)
to perform transcription of audio files to text (`Documents`). Please
see the notebook for usage.
Zep now supports persisting custom metadata with messages and hybrid
search across both message embeddings and structured metadata. This PR
implements custom metadata and enhancements to the
`ZepChatMessageHistory` and `ZepRetriever` classes to implement this
support.
Tag maintainers/contributors who might be interested:
VectorStores / Retrievers / Memory
- @dev2049
---------
Co-authored-by: Daniel Chalef <daniel.chalef@private.org>
# Check if generated Cypher code is wrapped in backticks
Some LLMs like the VertexAI like to explain how they generated the
Cypher statement and wrap the actual code in three backticks:
I have observed a similar pattern with OpenAI chat models in a
conversational settings, where multiple user and assistant message are
provided to the LLM to generate Cypher statements, where then the LLM
wants to maybe apologize for previous steps or explain its thoughts.
Interestingly, both OpenAI and VertexAI wrap the code in three backticks
if they are doing any explaining or apologizing. Checking if the
generated cypher is wrapped in backticks seems like a low-hanging fruit
to expand the cypher search to other LLMs and conversational settings.
# Token text splitter for sentence transformers
The current TokenTextSplitter only works with OpenAi models via the
`tiktoken` package. This is not clear from the name `TokenTextSplitter`.
In this (first PR) a token based text splitter for sentence transformer
models is added. In the future I think we should work towards injecting
a tokenizer into the TokenTextSplitter to make ti more flexible.
Could perhaps be reviewed by @dev2049
---------
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
Raises exception if OutputParsers receive a response with both a valid
action and a final answer
Currently, if an OutputParser receives a response which includes both an
action and a final answer, they return a FinalAnswer object. This allows
the parser to accept responses which propose an action and hallucinate
an answer without the action being parsed or taken by the agent.
This PR changes the logic to:
1. store a variable checking whether a response contains the
`FINAL_ANSWER_ACTION` (this is the easier condition to check).
2. store a variable checking whether the response contains a valid
action
3. if both are present, raise a new exception stating that both are
present
4. if an action is present, return an AgentAction
5. if an answer is present, return an AgentAnswer
6. if neither is present, raise the relevant exception based around the
action format (these have been kept consistent with the prior exception
messages)
Disclaimer:
* Existing mock data included strings which did include an action and an
answer. This might indicate that prioritising returning AgentAnswer was
always correct, and I am patching out desired behaviour? @hwchase17 to
advice. Curious if there are allowed cases where this is not
hallucinating, and we do want the LLM to output an action which isn't
taken.
* I have not passed `send_to_llm` through this new exception
Fixes#5601
## Who can review?
Community members can review the PR once tests pass. Tag
maintainers/contributors who might be interested:
@hwchase17 - project lead
@vowelparrot
# Unstructured Excel Loader
Adds an `UnstructuredExcelLoader` class for `.xlsx` and `.xls` files.
Works with `unstructured>=0.6.7`. A plain text representation of the
Excel file will be available under the `page_content` attribute in the
doc. If you use the loader in `"elements"` mode, an HTML representation
of the Excel file will be available under the `text_as_html` metadata
key. Each sheet in the Excel document is its own document.
### Testing
```python
from langchain.document_loaders import UnstructuredExcelLoader
loader = UnstructuredExcelLoader(
"example_data/stanley-cups.xlsx",
mode="elements"
)
docs = loader.load()
```
## Who can review?
@hwchase17
@eyurtsev
# Chroma update_document full document embeddings bugfix
Chroma update_document takes a single document, but treats the
page_content sting of that document as a list when getting the new
document embedding.
This is a two-fold problem, where the resulting embedding for the
updated document is incorrect (it's only an embedding of the first
character in the new page_content) and it calls the embedding function
for every character in the new page_content string, using many tokens in
the process.
Fixes#5582
Co-authored-by: Caleb Ellington <calebellington@Calebs-MBP.hsd1.ca.comcast.net>
# Fix Qdrant ids creation
There has been a bug in how the ids were created in the Qdrant vector
store. They were previously calculated based on the texts. However,
there are some scenarios in which two documents may have the same piece
of text but different metadata, and that's a valid case. Deduplication
should be done outside of insertion.
It has been fixed and covered with the integration tests.
---------
Co-authored-by: Dev 2049 <dev.dev2049@gmail.com>
# Fixes SQLAlchemy truncating the result if you have a big/text column
with many chars.
SQLAlchemy truncates columns if you try to convert a Row or Sequence to
a string directly
For comparison:
- Before:
```[('Harrison', 'That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio ... (2 characters truncated) ... hat is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio That is my Bio ')]```
- After:
```[('Harrison', 'That is my Bio That is my Bio That is my Bio That is
my Bio That is my Bio That is my Bio That is my Bio That is my Bio That
is my Bio That is my Bio That is my Bio That is my Bio That is my Bio
That is my Bio That is my Bio That is my Bio That is my Bio That is my
Bio That is my Bio That is my Bio ')]```
## Who can review?
Community members can review the PR once tests pass. Tag
maintainers/contributors who might be interested:
I'm not sure who to tag for chains, maybe @vowelparrot ?
Similar to #1813 for faiss, this PR is to extend functionality to pass
text and its vector pair to initialize and add embeddings to the
PGVector wrapper.
Community members can review the PR once tests pass. Tag
maintainers/contributors who might be interested:
- @dev2049
# Support Qdrant filters
Qdrant has an [extensive filtering
system](https://qdrant.tech/documentation/concepts/filtering/) with rich
type support. This PR makes it possible to use the filters in Langchain
by passing an additional param to both the
`similarity_search_with_score` and `similarity_search` methods.
## Who can review?
@dev2049 @hwchase17
---------
Co-authored-by: Dev 2049 <dev.dev2049@gmail.com>
when the LLMs output 'yes|no'，BooleanOutputParser can parse it to
'True|False', fix the ValueError in parse().
<!--
when use the BooleanOutputParser in the chain_filter.py, the LLMs output
'yes|no'，the function 'parse' will throw ValueError。
-->
Fixes # (issue)
#5396https://github.com/hwchase17/langchain/issues/5396
---------
Co-authored-by: gaofeng27692 <gaofeng27692@hundsun.com>
# Add maximal relevance search to SKLearnVectorStore
This PR implements the maximum relevance search in SKLearnVectorStore.
Twitter handle: jtolgyesi (I submitted also the original implementation
of SKLearnVectorStore)
## Before submitting
Unit tests are included.
Co-authored-by: Dev 2049 <dev.dev2049@gmail.com>
# Add batching to Qdrant
Several people requested a batching mechanism while uploading data to
Qdrant. It is important, as there are some limits for the maximum size
of the request payload, and without batching implemented in Langchain,
users need to implement it on their own. This PR exposes a new optional
`batch_size` parameter, so all the documents/texts are loaded in batches
of the expected size (64, by default).
The integration tests of Qdrant are extended to cover two cases:
1. Documents are sent in separate batches.
2. All the documents are sent in a single request.
# Handles the edge scenario in which the action input is a well formed
SQL query which ends with a quoted column
There may be a cleaner option here (or indeed other edge scenarios) but
this seems to robustly determine if the action input is likely to be a
well formed SQL query in which we don't want to arbitrarily trim off `"`
characters
Fixes#5423
## Who can review?
Community members can review the PR once tests pass. Tag
maintainers/contributors who might be interested:
For a quicker response, figure out the right person to tag with @
@hwchase17 - project lead
Agents / Tools / Toolkits
- @vowelparrot
# What does this PR do?
Bring support of `encode_kwargs` for ` HuggingFaceInstructEmbeddings`,
change the docstring example and add a test to illustrate with
`normalize_embeddings`.
Fixes#3605
(Similar to #3914)
Use case:
```python
from langchain.embeddings import HuggingFaceInstructEmbeddings
model_name = "hkunlp/instructor-large"
model_kwargs = {'device': 'cpu'}
encode_kwargs = {'normalize_embeddings': True}
hf = HuggingFaceInstructEmbeddings(
model_name=model_name,
model_kwargs=model_kwargs,
encode_kwargs=encode_kwargs
)
```
As the title says, I added more code splitters.
The implementation is trivial, so i don't add separate tests for each
splitter.
Let me know if any concerns.
Fixes # (issue)
https://github.com/hwchase17/langchain/issues/5170
## Who can review?
Community members can review the PR once tests pass. Tag
maintainers/contributors who might be interested:
@eyurtsev @hwchase17
---------
Signed-off-by: byhsu <byhsu@linkedin.com>
Co-authored-by: byhsu <byhsu@linkedin.com>
