Adds a cookbook for semi-structured RAG via Docugami. This follows the
same outline as the semi-structured RAG with Unstructured cookbook:
https://github.com/langchain-ai/langchain/blob/master/cookbook/Semi_Structured_RAG.ipynb
The main change is this cookbook uses Docugami instead of Unstructured
to find text and tables, and shows how XML markup in the output helps
with retrieval and generation.
We are \@docugami on twitter, I am \@tjaffri
---------
Co-authored-by: Taqi Jaffri <tjaffri@docugami.com>
- **Description:** We need to update the Dockerfile for templates to
also copy your README.md. This is because poetry requires that a readme
exists if it is specified in the pyproject.toml
Changes:
- remove langchain_core/schema since no clear distinction b/n schema and
non-schema modules
- make every module that doesn't end in -y plural
- where easy have 1-2 classes per file
- no more than one level of nesting in directories
- only import from top level core modules in langchain
<!-- Thank you for contributing to LangChain!
Replace this entire comment with:
- **Description:** a description of the change,
- **Issue:** the issue # it fixes (if applicable),
- **Dependencies:** any dependencies required for this change,
- **Tag maintainer:** for a quicker response, tag the relevant
maintainer (see below),
- **Twitter handle:** we announce bigger features on Twitter. If your PR
gets announced, and you'd like a mention, we'll gladly shout you out!
Please make sure your PR is passing linting and testing before
submitting. Run `make format`, `make lint` and `make test` to check this
locally.
See contribution guidelines for more information on how to write/run
tests, lint, etc:
https://github.com/langchain-ai/langchain/blob/master/.github/CONTRIBUTING.md
If you're adding a new integration, please include:
1. a test for the integration, preferably unit tests that do not rely on
network access,
2. an example notebook showing its use. It lives in `docs/extras`
directory.
If no one reviews your PR within a few days, please @-mention one of
@baskaryan, @eyurtsev, @hwchase17.
-->
- **Description:** fix a bug that prevented as_retriever() in Vectara to
use the desired input arguments
- **Issue:** as_retriever did not pass the arguments properly
- **Tag maintainer:** @baskaryan
- **Twitter handle:** @ofermend
I encountered this during summarization with VertexAI. I was receiving
an INVALID_ARGUMENT error, as it was trying to send a list of about
17000 single characters.
The [count_tokens
method](https://github.com/googleapis/python-aiplatform/blob/main/vertexai/language_models/_language_models.py#L658)
made available by Google takes in a list of prompts. It does not fail
for small texts, but it does for longer documents because the argument
list will be exceeding Googles allowed limit. Enforcing the list type
makes it work successfully.
This change will cast the input text to count to a list of that single
text so that the input format is always correct.
[Twitter](https://www.x.com/stijn_tratsaert)
- **Description:** ERNIE-Bot-Chat-4 Large Language Model adds the
ability of `Function Calling` by passing parameters through the
`functions` parameter in the request. To simplify function calling for
ERNIE-Bot-Chat-4, the `create_ernie_fn_chain()` function has been added.
The definition and usage of the `create_ernie_fn_chain()` function is
similar to that of the `create_openai_fn_chain()` function.
Examples as the follows:
```
import json
from langchain.chains.ernie_functions import (
create_ernie_fn_chain,
)
from langchain.chat_models import ErnieBotChat
from langchain.prompts import ChatPromptTemplate
def get_current_news(location: str) -> str:
"""Get the current news based on the location.'
Args:
location (str): The location to query.
Returs:
str: Current news based on the location.
"""
news_info = {
"location": location,
"news": [
"I have a Book.",
"It's a nice day, today."
]
}
return json.dumps(news_info)
def get_current_weather(location: str, unit: str="celsius") -> str:
"""Get the current weather in a given location
Args:
location (str): location of the weather.
unit (str): unit of the tempuature.
Returns:
str: weather in the given location.
"""
weather_info = {
"location": location,
"temperature": "27",
"unit": unit,
"forecast": ["sunny", "windy"],
}
return json.dumps(weather_info)
llm = ErnieBotChat(model_name="ERNIE-Bot-4")
prompt = ChatPromptTemplate.from_messages(
[
("human", "{query}"),
]
)
chain = create_ernie_fn_chain([get_current_weather, get_current_news], llm, prompt, verbose=True)
res = chain.run("北京今天的新闻是什么？")
print(res)
```
The running results of the above program are shown below：
```
> Entering new LLMChain chain...
Prompt after formatting:
Human: 北京今天的新闻是什么？
> Finished chain.
{'name': 'get_current_news', 'thoughts': '用户想要知道北京今天的新闻。我可以使用get_current_news工具来获取这些信息。', 'arguments': {'location': '北京'}}
```
- **Description:** during search with DeepLake some people are facing
backwards compatibility issues, this PR fixes it by making search
accessible for the older datasets
---------
Co-authored-by: adolkhan <adilkhan.sarsen@alumni.nu.edu.kz>
- **Description:**
- Fixes a `key_prefix` bug where passing it in on
`Redis.from_existing(...)` did not work properly. Updates doc strings
accordingly.
- Updates Redis filter classes logic with best practices on typing,
string formatting, and handling "empty" filters.
- Fixes a bug that would prevent multiple tag filters from being applied
together in some scenarios.
- Added a whole new filter unit testing module. Also updated code
formatting for a number of modules that were failing the `make`
commands.
- **Issue:** N/A
- **Dependencies:** N/A
- **Tag maintainer:** @baskaryan
- **Twitter handle:** @tchutch94
- **Description:** Fix typo in MongoDB memory docs
- **Tag maintainer:** @eyurtsev
<!-- Thank you for contributing to LangChain!
- **Description:** Fix typo in MongoDB memory docs
- **Issue:** the issue # it fixes (if applicable),
- **Dependencies:** any dependencies required for this change,
- **Tag maintainer:** @baskaryan
- **Twitter handle:** we announce bigger features on Twitter. If your PR
gets announced, and you'd like a mention, we'll gladly shout you out!
Please make sure your PR is passing linting and testing before
submitting. Run `make format`, `make lint` and `make test` to check this
locally.
See contribution guidelines for more information on how to write/run
tests, lint, etc:
https://github.com/langchain-ai/langchain/blob/master/.github/CONTRIBUTING.md
If you're adding a new integration, please include:
1. a test for the integration, preferably unit tests that do not rely on
network access,
2. an example notebook showing its use. It lives in `docs/extras`
directory.
If no one reviews your PR within a few days, please @-mention one of
@baskaryan, @eyurtsev, @hwchase17.
-->
In the `FORMAT_INSTRUCTIONS` template, 4 curly braces (escaping) are
used to get single curly brace after formatting:
```
"{{{ ... }}}}" -> format_instructions.format() -> "{{ ... }}" -> template.format() -> "{ ... }".
```
Tool's `args_schema` string contains single braces `{ ... }`, and is
also transformed to `{{{{ ... }}}}` form. But this is not really correct
since there is only one `format()` call:
```
"{{{{ ... }}}}" -> template.format() -> "{{ ... }}".
```
As a result we get double curly braces in the prompt:
````
Respond to the human as helpfully and accurately as possible. You have access to the following tools:
foo: Test tool FOO, args: {{'tool_input': {{'type': 'string'}}}} # <--- !!!
...
Provide only ONE action per $JSON_BLOB, as shown:
```
{
"action": $TOOL_NAME,
"action_input": $INPUT
}
```
````
This PR fixes curly braces escaping in the `args_schema` to have single
braces in the final prompt:
````
Respond to the human as helpfully and accurately as possible. You have access to the following tools:
foo: Test tool FOO, args: {'tool_input': {'type': 'string'}} # <--- !!!
...
Provide only ONE action per $JSON_BLOB, as shown:
```
{
"action": $TOOL_NAME,
"action_input": $INPUT
}
```
````
---------
Co-authored-by: Sergey Kozlov <sergey.kozlov@ludditelabs.io>
Hi 👋 We are working with Llama2 on Bedrock, and would like to add it to
Langchain. We saw a [pull
request](https://github.com/langchain-ai/langchain/pull/13322) to add it
to the `llm.Bedrock` class, but since it concerns a chat model, we would
like to add it to `BedrockChat` as well.
- **Description:** Add support for Llama2 to `BedrockChat` in
`chat_models`
- **Issue:** the issue # it fixes (if applicable)
[#13316](https://github.com/langchain-ai/langchain/issues/13316)
- **Dependencies:** any dependencies required for this change `None`
- **Tag maintainer:** /
- **Twitter handle:** `@SimonBockaert @WouterDurnez`
---------
Co-authored-by: wouter.durnez <wouter.durnez@showpad.com>
Co-authored-by: Simon Bockaert <simon.bockaert@showpad.com>
- **Description:** This change adds an agent to the Azure Cognitive
Services toolkit for identifying healthcare entities
- **Dependencies:** azure-ai-textanalytics (Optional)
---------
Co-authored-by: James Beck <James.Beck@sa.gov.au>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Hi!
This short PR aims at:
* Fixing `OpenAIEmbeddings`' check on `chunk_size` when used with Azure
OpenAI (thus with openai < 1.0). Azure OpenAI embeddings support at most
16 chunks per batch, I believe we are supposed to take the min between
the passed value/default value and 16, not the max - which, I suppose,
was introduced by accident while refactoring the previous version of
this check from this other PR of mine: #10707
* Porting this fix to the newest class (`AzureOpenAIEmbeddings`) for
openai >= 1.0
This fixes#13539 (closed but the issue persists).
@baskaryan @hwchase17
- **Description:** There are several mistakes in the sample code in the
doc-string of `DashVector` class, and this pull request aims to correct
them.
The correction code has been tested against latest version (at the time
of creation of this pull request) of: `langchain==0.0.336`
`dashvector==1.0.6` .
- **Issue:** No issue is created for this.
- **Dependencies:** No dependency is required for this change,
<!-- - **Tag maintainer:** for a quicker response, tag the relevant
maintainer (see below), -->
- **Twitter handle:** `zeyanglin`
<!-- Please make sure your PR is passing linting and testing before
submitting. Run `make format`, `make lint` and `make test` to check this
locally.
See contribution guidelines for more information on how to write/run
tests, lint, etc:
https://github.com/langchain-ai/langchain/blob/master/.github/CONTRIBUTING.md
If you're adding a new integration, please include:
1. a test for the integration, preferably unit tests that do not rely on
network access,
2. an example notebook showing its use. It lives in `docs/extras`
directory.
If no one reviews your PR within a few days, please @-mention one of
@baskaryan, @eyurtsev, @hwchase17.
-->
- **Description:** AstraDB is going to deprecate the `$similarity`
projection property in favor of the ´includeSimilarity´ option flag. I
moved all the queries to the new format.
- **Tag maintainer:** @hemidactylus
- **Twitter handle:** nicoloboschi
Added a `search_kwargs` field to BingSearchAPIWrapper in
`bing_search.py,` enabling users to include extra keyword arguments in
Bing search queries. This update, like specifying language preferences,
adds more customization to searches. The `search_kwargs` seamlessly
merge with standard parameters in `_bing_search_results` method.
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
- **Description:** Fix Astra integration tests that are failing. The
`delete` always return True as the deletion is successful if no errors
are thrown. I aligned the test to verify this behaviour
- **Tag maintainer:** @hemidactylus
- **Twitter handle:** nicoloboschi
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
The issue was accuring because of `openai` update in Completions. its
not accepting `api_key` and 'api_base' args.
The fix is we check for the openai version and if ats v1 then remove
these keys from args before passing them to `Compilation.create(...)`
when sending from `VLLMOpenAI`
Fixed: #13507
@eyu
@efriis
@hwchase17
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
- **Description:** In this pull request, we address an issue related to
assigning a schema to the SQLDatabase class when utilizing an Oracle
database. The current implementation encounters a bug where, upon
attempting to execute a query, the alter session parse is not
appropriately defined for Oracle, leading to an error,
- **Issue:** #7928,
- **Dependencies:** No dependencies,
- **Tag maintainer:** @baskaryan,
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
- **Description:** This change allows for the `MWDumpLoader` to load all
namespaces including custom by default instead of only loading the
[default
namespaces](https://www.mediawiki.org/wiki/Help:Namespaces#Localisation).
- **Tag maintainer:** @hwchase17
**Description:**
This commit adds embedchain retriever along with tests and docs.
Embedchain is a RAG framework to create data pipelines.
**Twitter handle:**
- [Taranjeet's twitter](https://twitter.com/taranjeetio) and
[Embedchain's twitter](https://twitter.com/embedchain)
**Reviewer**
@hwchase17
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
**Description:**
Enhance the functionality of YoutubeLoader to enable the translation of
available transcripts by refining the existing logic.
**Issue:**
Encountering a problem with YoutubeLoader (#13523) where the translation
feature is not functioning as expected.
Tag maintainers/contributors who might be interested:
@eyurtsev
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
BUG: langchain.agents.openai_assistant has a reference as
`from langchain_experimental.openai_assistant.base import
OpenAIAssistantRunnable`
should be
`from langchain.agents.openai_assistant.base import
OpenAIAssistantRunnable`
This prevents building of the API Reference docs
## Update 2023-09-08
This PR now supports further models in addition to Lllama-2 chat models.
See [this comment](#issuecomment-1668988543) for further details. The
title of this PR has been updated accordingly.
## Original PR description
This PR adds a generic `Llama2Chat` model, a wrapper for LLMs able to
serve Llama-2 chat models (like `LlamaCPP`,
`HuggingFaceTextGenInference`, ...). It implements `BaseChatModel`,
converts a list of chat messages into the [required Llama-2 chat prompt
format](https://huggingface.co/blog/llama2#how-to-prompt-llama-2) and
forwards the formatted prompt as `str` to the wrapped `LLM`. Usage
example:
```python
# uses a locally hosted Llama2 chat model
llm = HuggingFaceTextGenInference(
inference_server_url="http://127.0.0.1:8080/",
max_new_tokens=512,
top_k=50,
temperature=0.1,
repetition_penalty=1.03,
)
# Wrap llm to support Llama2 chat prompt format.
# Resulting model is a chat model
model = Llama2Chat(llm=llm)
messages = [
SystemMessage(content="You are a helpful assistant."),
MessagesPlaceholder(variable_name="chat_history"),
HumanMessagePromptTemplate.from_template("{text}"),
]
prompt = ChatPromptTemplate.from_messages(messages)
memory = ConversationBufferMemory(memory_key="chat_history", return_messages=True)
chain = LLMChain(llm=model, prompt=prompt, memory=memory)
# use chat model in a conversation
# ...
```
Also part of this PR are tests and a demo notebook.
- Tag maintainer: @hwchase17
- Twitter handle: `@mrt1nz`
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
- **Description:** Added a method `fetch_valid_documents` to
`WebResearchRetriever` class that will test the connection for every url
in `new_urls` and remove those that raise a `ConnectionError`.
- **Issue:** [Previous
PR](https://github.com/langchain-ai/langchain/pull/13353),
- **Dependencies:** None,
- **Tag maintainer:** @efriis
Please make sure your PR is passing linting and testing before
submitting. Run `make format`, `make lint` and `make test` to check this
locally.
See contribution guidelines for more information on how to write/run
tests, lint, etc:
https://github.com/langchain-ai/langchain/blob/master/.github/CONTRIBUTING.md
If you're adding a new integration, please include:
1. a test for the integration, preferably unit tests that do not rely on
network access,
2. an example notebook showing its use. It lives in `docs/extras`
directory.
If no one reviews your PR within a few days, please @-mention one of
@baskaryan, @eyurtsev, @hwchase17.
## Description
This PR adds an option to allow unsigned requests to the Neptune
database when using the `NeptuneGraph` class.
```python
graph = NeptuneGraph(
host='<my-cluster>',
port=8182,
sign=False
)
```
Also, added is an option in the `NeptuneOpenCypherQAChain` to provide
additional domain instructions to the graph query generation prompt.
This will be injected in the prompt as-is, so you should include any
provider specific tags, for example `<instructions>` or `<INSTR>`.
```python
chain = NeptuneOpenCypherQAChain.from_llm(
llm=llm,
graph=graph,
extra_instructions="""
Follow these instructions to build the query:
1. Countries contain airports, not the other way around
2. Use the airport code for identifying airports
"""
)
```
