- [x] NatbotChain: move to community, deprecate langchain version.
Update to use `prompt | llm | output_parser` instead of LLMChain.
- [x] LLMMathChain: deprecate + add langgraph replacement example to API
ref
- [x] HypotheticalDocumentEmbedder (retriever): update to use `prompt |
llm | output_parser` instead of LLMChain
- [x] FlareChain: update to use `prompt | llm | output_parser` instead
of LLMChain
- [x] ConstitutionalChain: deprecate + add langgraph replacement example
to API ref
- [x] LLMChainExtractor (document compressor): update to use `prompt |
llm | output_parser` instead of LLMChain
- [x] LLMChainFilter (document compressor): update to use `prompt | llm
| output_parser` instead of LLMChain
- [x] RePhraseQueryRetriever (retriever): update to use `prompt | llm |
output_parser` instead of LLMChain
Within the semantic chunker, when calling `_threshold_from_clusters`
there is the possibility for a divide by 0 error if the
`number_of_chunks` is equal to the length of `distances`.
Fix simply implements a check if these values match to prevent the error
and enable chunking to continue.
Remove the period after the hyperlink in the docstring of
BaseChatOpenAI.with_structured_output.
I have repeatedly copied the extra period at the end of the hyperlink,
which results in a "Page not found" page when pasted into the browser.
Fix typo
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Fix typo and some `callout` tags
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Backwards compatible change that converts pydantic extras to literals
which is consistent with pydantic 2 usage.
- fireworks
- voyage ai
- mistralai
- mistral ai
- together ai
- huggigng face
- pinecone
**Description**
Fix the asyncronous methods to retrieve documents from AzureSearch
VectorStore. The previous changes from [this
commit](ffe6ca986e)
create a similar code for the syncronous methods and the asyncronous
ones but the asyncronous client return an asyncronous iterator
"AsyncSearchItemPaged" as said in the issue #24740.
To solve this issue, the syncronous iterators in asyncronous methods
where changed to asyncronous iterators.
@chrislrobert said in [this
comment](https://github.com/langchain-ai/langchain/issues/24740#issuecomment-2254168302)
that there was a still a flaw due to `with` blocks that close the client
after each call. I removed this `with` blocks in the `async_client`
following the same pattern as the sync `client`.
In order to close up the connections, a __del__ method is included to
gently close up clients once the vectorstore object is destroyed.
**Issue:** #24740 and #24064
**Dependencies:** No new dependencies for this change
**Example notebook:** I created a notebook just to test the changes work
and gives the same results as the syncronous methods for vector and
hybrid search. With these changes, the asyncronous methods in the
retriever work as well.
**Lint and test**: Passes the tests and the linter
This adds `args_schema` member to `SearxSearchResults` tool. This member
is already present in the `SearxSearchRun` tool in the same file.
I was having `TypeError: Type is not JSON serializable:
AsyncCallbackManagerForToolRun` being thrown in langserve playground
when I was using `SearxSearchResults` tool as a part of chain there.
This fixes the issue, so the error is not raised anymore.
This is a example langserve app that was giving me the error, but it
works properly after the proposed fix:
```python
#!/usr/bin/env python
from fastapi import FastAPI
from langchain_core.prompts import ChatPromptTemplate
from langchain_core.output_parsers import StrOutputParser
from langchain_core.runnables import RunnablePassthrough
from langchain_openai import ChatOpenAI
from langchain_community.utilities import SearxSearchWrapper
from langchain_community.tools.searx_search.tool import SearxSearchResults
from langserve import add_routes
template = """Answer the question based only on the following context:
{context}
Question: {question}
"""
prompt = ChatPromptTemplate.from_template(template)
model = ChatOpenAI()
s = SearxSearchWrapper(searx_host="http://localhost:8080")
search = SearxSearchResults(wrapper=s)
search_chain = (
{"context": search, "question": RunnablePassthrough()}
| prompt
| model
| StrOutputParser()
)
app = FastAPI()
add_routes(
app,
search_chain,
path="/chain",
)
if __name__ == "__main__":
import uvicorn
uvicorn.run(app, host="localhost", port=8000)
```
- **Description:** Runhouse recently migrated from Read the Docs to a
self-hosted solution. This PR updates a broken link from the old docs to
www.run.house/docs. Also changed "The Runhouse" to "Runhouse" (it's
cleaner).
- **Issue:** None
- **Dependencies:** None
- **Description:** Standardize SparkLLM, include:
- docs, the issue #24803
- to support stream
- update api url
- model init arg names, the issue #20085
Cleaned up the "Tying it Together" section of the Conversational RAG
tutorial by removing unnecessary imports that were not used. This
reduces confusion and makes the code more concise.
Thank you for contributing to LangChain!
PR title: docs: remove unused imports in Conversational RAG tutorial
PR message:
Description: Removed unnecessary imports from the "Tying it Together"
section of the Conversational RAG tutorial. These imports were not used
in the code and created confusion. The updated code is now more concise
and easier to understand.
Issue: N/A
Dependencies: None
LinkedIn handle: [Hassan
Memon](https://www.linkedin.com/in/hassan-memon-a109b3257/)
Add tests and docs:
Hi [LangChain Team Member’s Name],
I hope you're doing well! I’m thrilled to share that I recently made my
second contribution to the LangChain project. If possible, could you
give me a shoutout on LinkedIn? It would mean a lot to me and could help
inspire others to contribute to the community as well.
Here’s my LinkedIn profile: [Hassan
Memon](https://www.linkedin.com/in/hassan-memon-a109b3257/).
Thank you so much for your support and for creating such a great
platform for learning and collaboration. I'm looking forward to
contributing more in the future!
Best regards,
Hassan Memon
fix: #25137
`SqliteSaver.from_conn_string()` has been changed to a `contextmanager`
method in `langgraph >= 0.2.0`, the original usage is no longer
applicable.
Refer to
<https://github.com/langchain-ai/langgraph/pull/1271#issue-2454736415>
modification method to replace `SqliteSaver` with `MemorySaver`.
- **Description:** This PR implements the `bind_tool` functionality for
ChatZhipuAI as requested by the user. ChatZhipuAI models support tool
calling according to the `OpenAI` tool format, as outlined in their
official documentation [here](https://open.bigmodel.cn/dev/api#glm-4).
- **Issue:** ##23868
---------
Co-authored-by: ccurme <chester.curme@gmail.com>
…ctions to match LangGraph v2 documentation. Corrected code snippet to
prevent validation errors.
Here's how you can fill out the provided template for your pull request:
---
**Thank you for contributing to LangChain!**
- [ ] **PR title**: `docs: update checkpointer example in Conversational
RAG tutorial`
- [ ] **PR message**:
- **Description:** Updated the Conversational RAG tutorial to correct
the checkpointer example by replacing `SqliteSaver` with `MemorySaver`.
Added installation instructions for `langgraph-checkpoint-memory` to
match LangGraph v2 documentation and prevent validation errors.
- **Issue:** N/A
- **Dependencies:** `langgraph-checkpoint-memory`
- **Twitter handle:** N/A
- [ ] **Add tests and docs**:
1. No new integration tests are required.
2. Updated documentation in the Conversational RAG tutorial.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: [LangChain Contribution
Guidelines](https://python.langchain.com/docs/contributing/)
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
**Description:** This PR rearranges the examples in Upstash Vector
integration documentation to describe how to use namespaces and improve
the description of metadata filtering.
Thank you for contributing to LangChain!
- **Description:** Fixing package install bug in cookbook
- **Issue:** zsh:1: no matches found: unstructured[all-docs]
- **Dependencies:** N/A
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
- **Description:** Fix link for API reference of Gmail Toolkit
- **Issue:** I've just found this issue while I'm reading the doc
- **Dependencies:** N/A
- **Twitter handle:** [@soichisumi](https://x.com/soichisumi)
TODO: If no one reviews your PR within a few days, please @-mention one
of baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
- In the in ` embedding-3 ` and later models of Zhipu AI, it is
supported to specify the dimensions parameter of Embedding. Ref:
https://bigmodel.cn/dev/api#text_embedding-3 .
- Add test case for `embedding-3` model by assigning dimensions.
This PR deprecates the beta upsert APIs in vectorstore.
We'll introduce them in a V2 abstraction instead to keep the existing
vectorstore implementations lighter weight.
The main problem with the existing APIs is that it's a bit more
challenging to
implement the correct behavior w/ respect to IDs since ID can be present
in
both the function signature and as an optional attribute on the document
object.
But VectorStores that pass the standard tests should have implemented
the semantics properly!
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
This PR gets rid `root_validators(allow_reuse=True)` logic used in
EdenAI Tool in preparation for pydantic 2 upgrade.
- add another test to secret_from_env_factory
**Description:**
The get time point method in the _consume() method of
core.rate_limiters.InMemoryRateLimiter uses time.time(), which can be
affected by system time backwards. Therefore, it is recommended to use
the monotonically increasing monotonic() to obtain the time
```python
with self._consume_lock:
now = time.time() # time.time() -> time.monotonic()
# initialize on first call to avoid a burst
if self.last is None:
self.last = now
elapsed = now - self.last # when use time.time(), elapsed may be negative when system time backwards
```
Thank you for contributing to LangChain!
- [X] **PR title**: "community: fix valueerror mentions wrong argument
missing"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [X] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** when faiss.py has a None relevance_score_fn it raises
a ValueError that says a normalize_fn_score argument is needed.
Co-authored-by: ccurme <chester.curme@gmail.com>
**Description:** This minor PR aims to add `llm_extraction` to Firecrawl
loader. This feature is supported on API and PythonSDK, but the
langchain loader omits adding this to the response.
**Twitter handle:** [scalable_pizza](https://x.com/scalablepizza)
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- Description: As described in the related issue: There is an error
occuring when using langchain-openai>=0.1.17 which can be attributed to
the following PR: #23691
Here, the parameter logprobs is added to requests per default.
However, AzureOpenAI takes issue with this parameter as stated here:
https://learn.microsoft.com/en-us/azure/ai-services/openai/how-to/chatgpt?tabs=python-new&pivots=programming-language-chat-completions
-> "If you set any of these parameters, you get an error."
Therefore, this PR changes the default value of logprobs parameter to
None instead of False. This results in it being filtered before the
request is sent.
- Issue: #24880
- Dependencies: /
Co-authored-by: blaufink <sebastian.brueckner@outlook.de>
Change all usages of __fields__ with get_fields adapter merged into
langchain_core.
Code mod generated using the following grit pattern:
```
engine marzano(0.1)
language python
`$X.__fields__` => `get_fields($X)` where {
add_import(source="langchain_core.utils.pydantic", name="get_fields")
}
```
Migrate pydantic extra to literals
Upgrade to using a literal for specifying the extra which is the
recommended approach in pydantic 2.
This works correctly also in pydantic v1.
```python
from pydantic.v1 import BaseModel
class Foo(BaseModel, extra="forbid"):
x: int
Foo(x=5, y=1)
```
And
```python
from pydantic.v1 import BaseModel
class Foo(BaseModel):
x: int
class Config:
extra = "forbid"
Foo(x=5, y=1)
```
## Enum -> literal using grit pattern:
```
engine marzano(0.1)
language python
or {
`extra=Extra.allow` => `extra="allow"`,
`extra=Extra.forbid` => `extra="forbid"`,
`extra=Extra.ignore` => `extra="ignore"`
}
```
Resorted attributes in config and removed doc-string in case we will
need to deal with going back and forth between pydantic v1 and v2 during
the 0.3 release. (This will reduce merge conflicts.)
## Sort attributes in Config:
```
engine marzano(0.1)
language python
function sort($values) js {
return $values.text.split(',').sort().join("\n");
}
class_definition($name, $body) as $C where {
$name <: `Config`,
$body <: block($statements),
$values = [],
$statements <: some bubble($values) assignment() as $A where {
$values += $A
},
$body => sort($values),
}
```
Add a utility that can be used as a default factory
The goal will be to start migrating from of the pydantic models to use
`from_env` as a default factory if possible.
```python
from pydantic import Field, BaseModel
from langchain_core.utils import from_env
class Foo(BaseModel):
name: str = Field(default_factory=from_env('HELLO'))
```
Change all usages of __fields__ with get_fields adapter merged into
langchain_core.
Code mod generated using the following grit pattern:
```
engine marzano(0.1)
language python
`$X.__fields__` => `get_fields($X)` where {
add_import(source="langchain_core.utils.pydantic", name="get_fields")
}
```
Upgrade to using a literal for specifying the extra which is the
recommended approach in pydantic 2.
This works correctly also in pydantic v1.
```python
from pydantic.v1 import BaseModel
class Foo(BaseModel, extra="forbid"):
x: int
Foo(x=5, y=1)
```
And
```python
from pydantic.v1 import BaseModel
class Foo(BaseModel):
x: int
class Config:
extra = "forbid"
Foo(x=5, y=1)
```
## Enum -> literal using grit pattern:
```
engine marzano(0.1)
language python
or {
`extra=Extra.allow` => `extra="allow"`,
`extra=Extra.forbid` => `extra="forbid"`,
`extra=Extra.ignore` => `extra="ignore"`
}
```
Resorted attributes in config and removed doc-string in case we will
need to deal with going back and forth between pydantic v1 and v2 during
the 0.3 release. (This will reduce merge conflicts.)
## Sort attributes in Config:
```
engine marzano(0.1)
language python
function sort($values) js {
return $values.text.split(',').sort().join("\n");
}
class_definition($name, $body) as $C where {
$name <: `Config`,
$body <: block($statements),
$values = [],
$statements <: some bubble($values) assignment() as $A where {
$values += $A
},
$body => sort($values),
}
```
Upgrade to using a literal for specifying the extra which is the
recommended approach in pydantic 2.
This works correctly also in pydantic v1.
```python
from pydantic.v1 import BaseModel
class Foo(BaseModel, extra="forbid"):
x: int
Foo(x=5, y=1)
```
And
```python
from pydantic.v1 import BaseModel
class Foo(BaseModel):
x: int
class Config:
extra = "forbid"
Foo(x=5, y=1)
```
## Enum -> literal using grit pattern:
```
engine marzano(0.1)
language python
or {
`extra=Extra.allow` => `extra="allow"`,
`extra=Extra.forbid` => `extra="forbid"`,
`extra=Extra.ignore` => `extra="ignore"`
}
```
Resorted attributes in config and removed doc-string in case we will
need to deal with going back and forth between pydantic v1 and v2 during
the 0.3 release. (This will reduce merge conflicts.)
## Sort attributes in Config:
```
engine marzano(0.1)
language python
function sort($values) js {
return $values.text.split(',').sort().join("\n");
}
class_definition($name, $body) as $C where {
$name <: `Config`,
$body <: block($statements),
$values = [],
$statements <: some bubble($values) assignment() as $A where {
$values += $A
},
$body => sort($values),
}
```
## Description
This PR adds back snippets demonstrating sparse and hybrid retrieval in
the Qdrant notebook.
Without the snippets, it's hard to grok the usage.
For business subscription the status is STOCKSBUSINESS not OK
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
## Description
This pull-request extends the existing vector search strategies of
MongoDBAtlasVectorSearch to include Hybrid (Reciprocal Rank Fusion) and
Full-text via new Retrievers.
There is a small breaking change in the form of the `prefilter` kwarg to
search. For this, and because we have now added a great deal of
features, including programmatic Index creation/deletion since 0.1.0, we
plan to bump the version to 0.2.0.
### Checklist
* Unit tests have been extended
* formatting has been applied
* One mypy error remains which will either go away in CI or be
simplified.
---------
Signed-off-by: Casey Clements <casey.clements@mongodb.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
Thank you for contributing to LangChain!
- [ ] **PR title**: "Documentation Update : Semantic Caching Update for
Upstash"
- Docs, llm caching integrations update
- **Description:** Upstash supports semantic caching, and we would like
to inform you about this
- **Twitter handle:** You can mention eray_eroglu_ if you want to post a
tweet about the PR
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- **Description:** Instantiating `GPT4AllEmbeddings` with no
`gpt4all_kwargs` argument raised a `ValidationError`. Root cause: #21238
added the capability to pass `gpt4all_kwargs` through to the `GPT4All`
instance via `Embed4All`, but broke code that did not specify a
`gpt4all_kwargs` argument.
- **Issue:** #25119
- **Dependencies:** None
- **Twitter handle:** [`@metadaddy`](https://twitter.com/metadaddy)
updated with langchain_google_community instead as the latest revision
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
This PR does an aesthetic sort of the config object attributes. This
will make it a bit easier to go back and forth between pydantic v1 and
pydantic v2 on the 0.3.x branch
Among integration packages in libs/partners, Groq is an exception in
that it errors on warnings.
Following https://github.com/langchain-ai/langchain/pull/25084, Groq
fails with
> pydantic.warnings.PydanticDeprecatedSince20: The `__fields__`
attribute is deprecated, use `model_fields` instead. Deprecated in
Pydantic V2.0 to be removed in V3.0.
Here we update the behavior to no longer fail on warning, which is
consistent with the rest of the packages in libs/partners.
**Description:**
In this PR, I am adding three stock market tools from
financialdatasets.ai (my API!):
- get balance sheets
- get cash flow statements
- get income statements
Twitter handle: [@virattt](https://twitter.com/virattt)
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Example: "community: Added bedrock 3-5 sonnet cost detials for
BedrockAnthropicTokenUsageCallbackHandler"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Co-authored-by: Naval Chand <navalchand@192.168.1.36>
- description: I remove the limitation of mandatory existence of
`QIANFAN_AK` and default model name which langchain uses cause there is
already a default model nama underlying `qianfan` SDK powering langchain
component.
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- community: Allow authorization to Confluence with bearer token
- **Description:** Allow authorization to Confluence with [Personal
Access
Token](https://confluence.atlassian.com/enterprise/using-personal-access-tokens-1026032365.html)
by checking for the keys `['client_id', token: ['access_token',
'token_type']]`
- **Issue:**
Currently the following error occurs when using an personal access token
for authorization.
```python
loader = ConfluenceLoader(
url=os.getenv('CONFLUENCE_URL'),
oauth2={
'token': {"access_token": os.getenv("CONFLUENCE_ACCESS_TOKEN"), "token_type": "bearer"},
'client_id': 'client_id',
},
page_ids=['12345678'],
)
```
```
ValueError: Error(s) while validating input: ["You have either omitted require keys or added extra keys to the oauth2 dictionary. key values should be `['access_token', 'access_token_secret', 'consumer_key', 'key_cert']`"]
```
With this PR the loader runs as expected.
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- **Description:** This includes Pydantic field metadata in
`_create_subset_model_v2` so that it gets included in the final
serialized form that get sent out.
- **Issue:** #25031
- **Dependencies:** n/a
- **Twitter handle:** @gramliu
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Fixes Neo4JVector.from_existing_graph integration with huggingface
Previously threw an error with existing databases, because
from_existing_graph query returns empty list of new nodes, which are
then passed to embedding function, and huggingface errors with empty
list.
Fixes [24401](https://github.com/langchain-ai/langchain/issues/24401)
---------
Co-authored-by: Jeff Katzy <jeffreyerickatz@gmail.com>
You can use this with:
```
from langchain_experimental.graph_transformers import GlinerGraphTransformer
gliner = GlinerGraphTransformer(allowed_nodes=["Person", "Organization", "Nobel"], allowed_relationships=["EMPLOYEE", "WON"])
from langchain_core.documents import Document
text = """
Marie Curie, was a Polish and naturalised-French physicist and chemist who conducted pioneering research on radioactivity.
She was the first woman to win a Nobel Prize, the first person to win a Nobel Prize twice, and the only person to win a Nobel Prize in two scientific fields.
Her husband, Pierre Curie, was a co-winner of her first Nobel Prize, making them the first-ever married couple to win the Nobel Prize and launching the Curie family legacy of five Nobel Prizes.
She was, in 1906, the first woman to become a professor at the University of Paris.
"""
documents = [Document(page_content=text)]
gliner.convert_to_graph_documents(documents)
```
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
This PR adds a minimal document indexer abstraction.
The goal of this abstraction is to allow developers to create custom
retrievers that also have a standard indexing API and allow updating the
document content in them.
The abstraction comes with a test suite that can verify that the indexer
implements the correct semantics.
This is an iteration over a previous PRs
(https://github.com/langchain-ai/langchain/pull/24364). The main
difference is that we're sub-classing from BaseRetriever in this
iteration and as so have consolidated the sync and async interfaces.
The main problem with the current design is that runt time search
configuration has to be specified at init rather than provided at run
time.
We will likely resolve this issue in one of the two ways:
(1) Define a method (`get_retriever`) that will allow creating a
retriever at run time with a specific configuration.. If we do this, we
will likely break the subclass on BaseRetriever
(2) Generalize base retriever so it can support structured queries
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
- [x] **PR title**: "docs: changed example for Exa search retriever
usage"
- [x] **PR message**:
- **Description:** Changed Exa integration doc at
`docs/docs/integrations/tools/exa_search.ipynb` to better reflect simple
Exa use case
- **Issue:** move toward more canonical use of Exa method
(`search_and_contents` rather than just `search`)
- **Dependencies:** no dependencies; docs only change
- **Twitter handle:** n/a - small change
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17. - will do
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
**Description:**
This PR fixes a bug where if `enable_dynamic_field` and
`partition_key_field` are enabled at the same time, a pymilvus error
occurs.
Milvus requires the partition key field to be a full schema defined
field, and not a dynamic one, so it will throw the error "the specified
partition key field {field} not exist" when creating the collection.
When `enabled_dynamic_field` is set to `True`, all schema field creation
based on `metadatas` is skipped. This code now checks if
`partition_key_field` is set, and creates the field.
Integration test added.
**Twitter handle:** StuartMarshUK
---------
Co-authored-by: Stuart Marsh <stuart.marsh@qumata.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
- **Description:** This PR makes the AthenaLoader profile_name optional
and fixes the type hint which says the type is `str` but it should be
`str` or `None` as None is handled in the loader init. This is a minor
problem but it just confused me when I was using the Athena Loader to
why we had to use a Profile, as I want that for local but not
production.
- **Issue:** #24957
- **Dependencies:** None.
Description: RetryWithErrorOutputParser.from_llm() creates a retry chain
that returns a Generation instance, when it should actually just return
a string.
This class was forgotten when fixing the issue in PR #24687
The comments inside some code blocks seems to be misplaced. The comment
lines containing explanation about `default_key` behavior when operating
with prompts are updated.
Thank you for contributing to LangChain!
- [x] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [x] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Added to `docs/how_to/tools_runtime` as a proof of concept, will apply
everywhere if we like.
A bit more compact than the default callouts, will help standardize the
layout of our pages since we frequently use these boxes.
<img width="1088" alt="Screenshot 2024-07-23 at 4 49 02 PM"
src="https://github.com/user-attachments/assets/7380801c-e092-4d31-bcd8-3652ee05f29e">
Hardens index commands with try/except for free clusters and optional
waits for syncing and tests.
[efriis](https://github.com/efriis) These are the upgrades to the search
index commands (CRUD) that I mentioned.
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
- **Description:** The UnstructuredClient will have a breaking change in
the near future. Add a note in the docs that the examples here may not
use the latest version and users should refer to the SDK docs for the
latest info.
This PR introduces a module with some helper utilities for the pydantic
1 -> 2 migration.
They're meant to be used in the following way:
1) Use the utility code to get unit tests pass without requiring
modification to the unit tests
2) (If desired) upgrade the unit tests to match pydantic 2 output
3) (If desired) stop using the utility code
Currently, this module contains a way to map `schema()` generated by
pydantic 2 to (mostly) match the output from pydantic v1.
Thank you for contributing to LangChain!
- [x] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- **Description:**
Support ChatMlflow.bind_tools method
Tested in Databricks:
<img width="836" alt="image"
src="https://github.com/user-attachments/assets/fa28ef50-0110-4698-8eda-4faf6f0b9ef8">
- [x] **Add tests and docs**: 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/docs/integrations` directory.
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
---------
Signed-off-by: Serena Ruan <serena.rxy@gmail.com>
- **Description:** When adding docs for constructing ChatHuggingFace
using a HuggingFacePipeline, I forgot to add `return_full_text=False` as
an argument. In this setup, the chat response would incorrectly contain
all the input text. I am fixing that here by adding that line to the
offending notebook.
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Thank you for contributing to LangChain!
- [x] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [x] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [x] **Add tests and docs**: 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/docs/integrations` directory.
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
**Description:** This PR fixes a KeyError in NotionDBLoader when the
"name" key is missing in the "people" property.
**Issue:** Fixes#24223
**Dependencies:** None
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
instead of hardcoding a linter for each, iterate through the lines of
the template notebook and find lines that start with `##` (includes
lower headings), and enforce that those headings are found in new docs
that are contributed
Add compatibility for pydantic 2 for a utility function.
This will help push some small changes to master, so they don't have to
be kept track of on a separate branch.
The @pre_init validator is a temporary solution for base models. It has
similar (but not identical) semantics to @root_validator(), but it works
strictly as a pre-init validator.
It'll work as expected as long as the pydantic model type hints were
correct.
supports following UX
```python
class SubTool(TypedDict):
"""Subtool docstring"""
args: Annotated[Dict[str, Any], {}, "this does bar"]
class Tool(TypedDict):
"""Docstring
Args:
arg1: foo
"""
arg1: str
arg2: Union[int, str]
arg3: Optional[List[SubTool]]
arg4: Annotated[Literal["bar", "baz"], ..., "this does foo"]
arg5: Annotated[Optional[float], None]
```
- can parse google style docstring
- can use Annotated to specify default value (second arg)
- can use Annotated to specify arg description (third arg)
- can have nested complex types
This PR adds annotations in comunity package.
Annotations are only strictly needed in subclasses of BaseModel for
pydantic 2 compatibility.
This PR adds some unnecessary annotations, but they're not bad to have
regardless for documentation pages.
Title: [pebblo_retrieval] Identifying entities in prompts given in
PebbloRetrievalQA leading to prompt governance
Description: Implemented identification of entities in the prompt using
Pebblo prompt governance API.
Issue: NA
Dependencies: NA
Add tests and docs: NA
- **Title:** [PebbloSafeLoader] Implement content-size-based batching in
the classification flow(loader/doc API)
- **Description:**
- Implemented content-size-based batching in the loader/doc API, set to
100KB with no external configuration option, intentionally hard-coded to
prevent timeouts.
- Remove unused field(pb_id) from doc_metadata
- **Issue:** NA
- **Dependencies:** NA
- **Add tests and docs:** Updated
Description: The old method will be discontinued; use the official SDK
for more model options.
Issue: None
Dependencies: None
Twitter handle: None
Co-authored-by: trumanyan <trumanyan@tencent.com>
**Description:** Updated the Langgraph migration docs to use
`state_modifier` rather than `messages_modifier`
**Issue:** N/A
**Dependencies:** N/A
- [ X] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
PR title: Experimental: Add config to convert_to_graph_documents
Description: In order to use langfuse, i need to pass the langfuse
configuration when invoking the chain. langchain_experimental does not
allow to add any parameters (beside the documents) to the
convert_to_graph_documents method. This way, I cannot monitor the chain
in langfuse.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
---------
Co-authored-by: Catarina Franco <catarina.franco@criticalsoftware.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
## Description
This PR:
- Fixes the validation error in `FastEmbedEmbeddings`.
- Adds support for `batch_size`, `parallel` params.
- Removes support for very old FastEmbed versions.
- Updates the FastEmbed doc with the new params.
Associated Issues:
- Resolves#24039
- Resolves #https://github.com/qdrant/fastembed/issues/296
**Description:**
This update significantly improves the Brave Search Tool's utility
within the LangChain library by enriching the search results it returns.
The tool previously returned title, link, and snippet, with the snippet
being a truncated 140-character description from the search engine. To
make the search results more informative, this update enables
extra_snippets by default and introduces additional result fields:
title, link, description (enhancing and renaming the former snippet
field), age, and snippets. The snippets field provides a list of strings
summarizing the webpage, utilizing Brave's capability for more detailed
search insights. This enhancement aims to make the search tool far more
informative and beneficial for users.
**Issue:** N/A
**Dependencies:** No additional dependencies introduced.
**Twitter handle:** @davidalexr987
**Code Changes Summary:**
- Changed the default setting to include extra_snippets in search
results.
- Renamed the snippet field to description to accurately reflect its
content and included an age field for search results.
- Introduced a snippets field that lists webpage summaries, providing
users with comprehensive search result insights.
**Backward Compatibility Note:**
The renaming of snippet to description improves the accuracy of the
returned data field but may impact existing users who have developed
integration's or analyses based on the snippet field. I believe this
change is essential for clarity and utility, and it aligns better with
the data provided by Brave Search.
**Additional Notes:**
This proposal focuses exclusively on the Brave Search package, without
affecting other LangChain packages or introducing new dependencies.
Description: Since moving away from `langchain-community` is
recommended, `init_chat_models()` should import ChatOllama from
`langchain-ollama` instead.
Anthropic models (including via Bedrock and other cloud platforms)
accept a status/is_error attribute on tool messages/results
(specifically in `tool_result` content blocks for Anthropic API). Adding
a ToolMessage.status attribute so that users can set this attribute when
using those models
**Description:** Add empty string default for api_key and change
`server_url` to `url` to match existing loaders.
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
**Description**
Fixes DocumentDBVectorSearch similarity_search when no filter is used;
it defaults to None but $match does not accept None, so changed default
to empty {} before pipeline is created.
**Issue**
AWS DocumentDB similarity search does not work when no filter is used.
Error msg: "the match filter must be an expression in an object" #24775
**Dependencies**
No dependencies
**Twitter handle**
https://x.com/perepasamonte
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- Mixtral with Groq has started consistently failing tool calling tests.
Here we restrict testing to llama 3.1.
- `.schema` is deprecated in pydantic proper in favor of
`.model_json_schema`.
There is an issue with the prompt format in `GenerativeAgentMemory` ,
try to fix it.
The prompt is same as the one in method `_score_memory_importance`.
issue: #24615
descriptions: The _Graph pydantic model generated from
create_simple_model (which LLMGraphTransformer uses when allowed nodes
and relationships are provided) does not constrain the relationships
(source and target types, relationship type), and the node and
relationship properties with enums when using ChatOpenAI.
The issue is that when calling optional_enum_field throughout
create_simple_model the llm_type parameter is not passed in except for
when creating node type. Passing it into each call fixes the issue.
Co-authored-by: Lifu Wu <lifu@nextbillion.ai>
- [ ] **PR title**: "langchain-openai: openai proxy added to base
embeddings"
- [ ] **PR message**:
- **Description:**
Dear langchain developers,
You've already supported proxy for ChatOpenAI implementation in your
package. At the same time, if somebody needed to use proxy for chat, it
also could be necessary to be able to use it for OpenAIEmbeddings.
That's why I think it's important to add proxy support for OpenAI
embeddings. That's what I've done in this PR.
@baskaryan
---------
Co-authored-by: karpov <karpov@dohod.ru>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Thank you for contributing to LangChain!
- [x] **PR title**: "Add documentaiton on InMemoryVectorStore driver for
MemoryDB to langchain-aws"
- Langchain-aws repo :Add MemoryDB documentation
- Example: "community: add foobar LLM"
- [x] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** Added documentation on InMemoryVectorStore driver to
aws.mdx and usage example on MemoryDB clusuter
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [x] **Add tests and docs**: If you're adding a new integration, please
include
Add memorydb notebook to docs/docs/integrations/ folde
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
**Description:**
In the `ChatFireworks` class definition, the Field() call for the "stop"
("stop_sequences") parameter is missing the "default" keyword.
**Issue:**
Type checker reports "stop_sequences" as a missing arg (not recognizing
the default value is None)
**Dependencies:**
None
**Twitter handle:**
None
Description: OutputFixingParser.from_llm() creates a retry chain that
returns a Generation instance, when it should actually just return a
string.
Issue: https://github.com/langchain-ai/langchain/issues/24600
Twitter handle: scribu
---------
Co-authored-by: isaac hershenson <ihershenson@hmc.edu>
Thank you for contributing to LangChain!
- [x] **PR title**: "community:add Yi LLM", "docs:add Yi Documentation"
- [x] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** This PR adds support for the Yi model to LangChain.
- **Dependencies:**
[langchain_core,requests,contextlib,typing,logging,json,langchain_community]
- **Twitter handle:** 01.AI
- [x] **Add tests and docs**: I've added the corresponding documentation
to the relevant paths
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Co-authored-by: isaac hershenson <ihershenson@hmc.edu>
Raise `LangChainException` instead of `Exception`. This alleviates the
need for library users to use bare try/except to handle exceptions
raised by `AzureSearch`.
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
Description:
add a optional score relevance threshold for select only coherent
document, it's in complement of top_n
Discussion:
add relevance score threshold in flashrank_rerank document compressors
#24013
Dependencies:
no dependencies
---------
Co-authored-by: Benjamin BERNARD <benjamin.bernard@openpathview.fr>
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Description:
- This PR adds a self query retriever implementation for SAP HANA Cloud
Vector Engine. The retriever supports all operators except for contains.
- Issue: N/A
- Dependencies: no new dependencies added
**Add tests and docs:**
Added integration tests to:
libs/community/tests/unit_tests/query_constructors/test_hanavector.py
**Documentation for self query retriever:**
/docs/integrations/retrievers/self_query/hanavector_self_query.ipynb
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
**Description:** Expanded the chat model functionality to support tools
in the 'baichuan.py' file. Updated module imports and added tool object
handling in message conversions. Additional changes include the
implementation of tool binding and related unit tests. The alterations
offer enhanced model capabilities by enabling interaction with tool-like
objects.
---------
Co-authored-by: ccurme <chester.curme@gmail.com>
- [x] **PR title**:
community: Add OCI Generative AI tool and structured output support
- [x] **PR message**:
- **Description:** adding tool calling and structured output support for
chat models offered by OCI Generative AI services. This is an update to
our last PR 22880 with changes in
/langchain_community/chat_models/oci_generative_ai.py
- **Issue:** NA
- **Dependencies:** NA
- **Twitter handle:** NA
- [x] **Add tests and docs**:
1. we have updated our unit tests
2. we have updated our documentation under
/docs/docs/integrations/chat/oci_generative_ai.ipynb
- [x] **Lint and test**: `make format`, `make lint` and `make test` we
run successfully
---------
Co-authored-by: RHARPAZ <RHARPAZ@RHARPAZ-5750.us.oracle.com>
Co-authored-by: Arthur Cheng <arthur.cheng@oracle.com>
This PR proposes to create a rate limiter in the chat model directly,
and would replace: https://github.com/langchain-ai/langchain/pull/21992
It resolves most of the constraints that the Runnable rate limiter
introduced:
1. It's not annoying to apply the rate limiter to existing code; i.e.,
possible to roll out the change at the location where the model is
instantiated,
rather than at every location where the model is used! (Which is
necessary
if the model is used in different ways in a given application.)
2. batch rate limiting is enforced properly
3. the rate limiter works correctly with streaming
4. the rate limiter is aware of the cache
5. The rate limiter can take into account information about the inputs
into the
model (we can add optional inputs to it down-the road together with
outputs!)
The only downside is that information will not be properly reflected in
tracing
as we don't have any metadata evens about a rate limiter. So the total
time
spent on a model invocation will be:
* time spent waiting for the rate limiter
* time spend on the actual model request
## Example
```python
from langchain_core.rate_limiters import InMemoryRateLimiter
from langchain_groq import ChatGroq
groq = ChatGroq(rate_limiter=InMemoryRateLimiter(check_every_n_seconds=1))
groq.invoke('hello')
```
**Description:**
- This PR exposes some functions in VDMS vectorstore, updates VDMS
related notebooks, updates tests, and upgrade version of VDMS (>=0.0.20)
**Issue:** N/A
**Dependencies:**
- Update vdms>=0.0.20
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Lots of duplicated content from concepts, missing pointers to the second
half of the tool calling loop
Simpler + more focused + a more prominent link to the second half of the
loop was what I was aiming for, but down to be more conservative and
just more prominently link the "passing tools back to the model" guide.
I have also moved the tool calling conceptual guide out from under
`Structured Output` (while leaving a small section for structured
output-specific information) and added more content. The existing
`#functiontool-calling` link will go to this new section.
Fixes for Eden AI Custom tools and ChatEdenAI:
- add missing import in __init__ of chat_models
- add `args_schema` to custom tools. otherwise '__arg1' would sometimes
be passed to the `run` method
- fix IndexError when no human msg is added in ChatEdenAI
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Mistral appears to have added validation for the format of its tool call
IDs:
`{"object":"error","message":"Tool call id was abc123 but must be a-z,
A-Z, 0-9, with a length of
9.","type":"invalid_request_error","param":null,"code":null}`
This breaks compatibility of messages from other providers. Here we add
a function that converts any string to a Mistral-valid tool call ID, and
apply it to incoming messages.
Thank you for contributing to LangChain!
**Description:**
This PR allows users of `langchain_community.llms.ollama.Ollama` to
specify the `auth` parameter, which is then forwarded to all internal
calls of `requests.request`. This works in the same way as the existing
`headers` parameters. The auth parameter enables the usage of the given
class with Ollama instances, which are secured by more complex
authentication mechanisms, that do not only rely on static headers. An
example are AWS API Gateways secured by the IAM authorizer, which
expects signatures dynamically calculated on the specific HTTP request.
**Issue:**
Integrating a remote LLM running through Ollama using
`langchain_community.llms.ollama.Ollama` only allows setting static HTTP
headers with the parameter `headers`. This does not work, if the given
instance of Ollama is secured with an authentication mechanism that
makes use of dynamically created HTTP headers which for example may
depend on the content of a given request.
**Dependencies:**
None
**Twitter handle:**
None
---------
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
### Description
* support asynchronous in InMemoryVectorStore
* since embeddings might be possible to call asynchronously, ensure that
both asynchronous and synchronous functions operate correctly.
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
This PR introduces the following Runnables:
1. BaseRateLimiter: an abstraction for specifying a time based rate
limiter as a Runnable
2. InMemoryRateLimiter: Provides an in-memory implementation of a rate
limiter
## Example
```python
from langchain_core.runnables import InMemoryRateLimiter, RunnableLambda
from datetime import datetime
foo = InMemoryRateLimiter(requests_per_second=0.5)
def meow(x):
print(datetime.now().strftime("%H:%M:%S.%f"))
return x
chain = foo | meow
for _ in range(10):
print(chain.invoke('hello'))
```
Produces:
```
17:12:07.530151
hello
17:12:09.537932
hello
17:12:11.548375
hello
17:12:13.558383
hello
17:12:15.568348
hello
17:12:17.578171
hello
17:12:19.587508
hello
17:12:21.597877
hello
17:12:23.607707
hello
17:12:25.617978
hello
```
## Interface
The rate limiter uses the following interface for acquiring a token:
```python
class BaseRateLimiter(Runnable[Input, Output], abc.ABC):
@abc.abstractmethod
def acquire(self, *, blocking: bool = True) -> bool:
"""Attempt to acquire the necessary tokens for the rate limiter.```
```
The flag `blocking` has been added to the abstraction to allow
supporting streaming (which is easier if blocking=False).
## Limitations
- The rate limiter is not designed to work across different processes.
It is an in-memory rate limiter, but it is thread safe.
- The rate limiter only supports time-based rate limiting. It does not
take into account the size of the request or any other factors.
- The current implementation does not handle streaming inputs well and
will consume all inputs even if the rate limit has been reached. Better
support for streaming inputs will be added in the future.
- When the rate limiter is combined with another runnable via a
RunnableSequence, usage of .batch() or .abatch() will only respect the
average rate limit. There will be bursty behavior as .batch() and
.abatch() wait for each step to complete before starting the next step.
One way to mitigate this is to use batch_as_completed() or
abatch_as_completed().
## Bursty behavior in `batch` and `abatch`
When the rate limiter is combined with another runnable via a
RunnableSequence, usage of .batch() or .abatch() will only respect the
average rate limit. There will be bursty behavior as .batch() and
.abatch() wait for each step to complete before starting the next step.
This becomes a problem if users are using `batch` and `abatch` with many
inputs (e.g., 100). In this case, there will be a burst of 100 inputs
into the batch of the rate limited runnable.
1. Using a RunnableBinding
The API would look like:
```python
from langchain_core.runnables import InMemoryRateLimiter, RunnableLambda
rate_limiter = InMemoryRateLimiter(requests_per_second=0.5)
def meow(x):
return x
rate_limited_meow = RunnableLambda(meow).with_rate_limiter(rate_limiter)
```
2. Another option is to add some init option to RunnableSequence that
changes `.batch()` to be depth first (e.g., by delegating to
`batch_as_completed`)
```python
RunnableSequence(first=rate_limiter, last=model, how='batch-depth-first')
```
Pros: Does not require Runnable Binding
Cons: Feels over-complicated
Added [ScrapingAnt](https://scrapingant.com/) Web Loader integration.
ScrapingAnt is a web scraping API that allows extracting web page data
into accessible and well-formatted markdown.
Description: Added ScrapingAnt web loader for retrieving web page data
as markdown
Dependencies: scrapingant-client
Twitter: @WeRunTheWorld3
---------
Co-authored-by: Oleg Kulyk <oleg@scrapingant.com>
#### Update (2):
A single `UnstructuredLoader` is added to handle both local and api
partitioning. This loader also handles single or multiple documents.
#### Changes in `community`:
Changes here do not affect users. In the initial process of using the
SDK for the API Loaders, the Loaders in community were refactored.
Other changes include:
The `UnstructuredBaseLoader` has a new check to see if both
`mode="paged"` and `chunking_strategy="by_page"`. It also now has
`Element.element_id` added to the `Document.metadata`.
`UnstructuredAPIFileLoader` and `UnstructuredAPIFileIOLoader`. As such,
now both directly inherit from `UnstructuredBaseLoader` and initialize
their `file_path`/`file` attributes respectively and implement their own
`_post_process_elements` methods.
--------
#### Update:
New SDK Loaders in a [partner
package](https://python.langchain.com/v0.1/docs/contributing/integrations/#partner-package-in-langchain-repo)
are introduced to prevent breaking changes for users (see discussion
below).
##### TODO:
- [x] Test docstring examples
--------
- **Description:** UnstructuredAPIFileIOLoader and
UnstructuredAPIFileLoader calls to the unstructured api are now made
using the unstructured-client sdk.
- **New Dependencies:** unstructured-client
- [x] **Add tests and docs**: If you're adding a new integration, please
include
- [x] a test for the integration, preferably unit tests that do not rely
on network access,
- [x] update the description in
`docs/docs/integrations/providers/unstructured.mdx`
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
TODO:
- [x] Update
https://python.langchain.com/v0.1/docs/integrations/document_loaders/unstructured_file/#unstructured-api
-
`langchain/docs/docs/integrations/document_loaders/unstructured_file.ipynb`
- The description here needs to indicate that users should install
`unstructured-client` instead of `unstructured`. Read over closely to
look for any other changes that need to be made.
- [x] Update the `lazy_load` method in `UnstructuredBaseLoader` to
handle json responses from the API instead of just lists of elements.
- This method may need to be overwritten by the API loaders instead of
changing it in the `UnstructuredBaseLoader`.
- [x] Update the documentation links in the class docstrings (the
Unstructured documents have moved)
- [x] Update Document.metadata to include `element_id` (see thread
[here](https://unstructuredw-kbe4326.slack.com/archives/C044N0YV08G/p1718187499818419))
---------
Signed-off-by: ChengZi <chen.zhang@zilliz.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
Co-authored-by: Isaac Francisco <78627776+isahers1@users.noreply.github.com>
Co-authored-by: ChengZi <chen.zhang@zilliz.com>
- [ ] **PR title**: "experimental: Adding compatibility for
OllamaFunctions with ImagePromptTemplate"
- [ ] **PR message**:
- **Description:** Removes the outdated
`_convert_messages_to_ollama_messages` method override in the
`OllamaFunctions` class to ensure that ollama multimodal models can be
invoked with an image.
- **Issue:** #24174
---------
Co-authored-by: Joel Akeret <joel.akeret@ti&m.com>
Co-authored-by: Isaac Francisco <78627776+isahers1@users.noreply.github.com>
Co-authored-by: isaac hershenson <ihershenson@hmc.edu>
add dynamic field feature to langchain_milvus
more unittest, more robustic
plan to deprecate the `metadata_field` in the future, because it's
function is the same as `enable_dynamic_field`, but the latter one is a
more advanced concept in milvus
Signed-off-by: ChengZi <chen.zhang@zilliz.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
This linter is meant to move development to use __init__ instead of
root_validator and validator.
We need to investigate whether we need to lint some of the functionality
of Field (e.g., `lt` and `gt`, `alias`)
`alias` is the one that's most popular:
(community) ➜ community git:(eugene/add_linter_to_community) ✗ git grep
" Field(" | grep "alias=" | wc -l
144
(community) ➜ community git:(eugene/add_linter_to_community) ✗ git grep
" Field(" | grep "ge=" | wc -l
10
(community) ➜ community git:(eugene/add_linter_to_community) ✗ git grep
" Field(" | grep "gt=" | wc -l
4
This PR is under WIP and adds the following functionalities:
- [X] Supports tool calling across the langchain ecosystem. (However
streaming is not supported)
- [X] Update documentation
- [ ] **Community**: "Retrievers: Product Quantization"
- [X] This PR adds Product Quantization feature to the retrievers to the
Langchain Community. PQ is one of the fastest retrieval methods if the
embeddings are rich enough in context due to the concepts of
quantization and representation through centroids
- **Description:** Adding PQ as one of the retrievers
- **Dependencies:** using the package nanopq for this PR
- **Twitter handle:** vishnunkumar_
- [X] **Add tests and docs**: If you're adding a new integration, please
include
- [X] Added unit tests for the same in the retrievers.
- [] Will add an example notebook subsequently
- [X] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/ -
done the same
---------
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Thank you for contributing to LangChain!
- [x] **PR title**: Update IBM docs about information to pass client
into WatsonxLLM and WatsonxEmbeddings object.
- [x] **PR message**:
- **Description:** Update IBM docs about information to pass client into
WatsonxLLM and WatsonxEmbeddings object.
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Thank you for contributing to LangChain!
- This PR adds vector search filtering for Azure Cosmos DB Mongo vCore
and NoSQL.
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
**Description**
Add support for Pinecone hosted embedding models as
`PineconeEmbeddings`. Replacement for #22890
**Dependencies**
Add `aiohttp` to support async embeddings call against REST directly
- [x] **Add tests and docs**: If you're adding a new integration, please
include
Added `docs/docs/integrations/text_embedding/pinecone.ipynb`
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Twitter: `gdjdg17`
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
In some lines its trying to read a key that do not exists yet. In this
cases I changed the direct access to dict.get() method
- [ x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
The previous implementation would never be called.
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, hwchase17.
---------
Co-authored-by: Isaac Francisco <78627776+isahers1@users.noreply.github.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
After this standard tests will test with the following combinations:
1. pydantic.BaseModel
2. pydantic.v1.BaseModel
If ran within a matrix, it'll covert both pydantic.BaseModel originating
from
pydantic 1 and the one defined in pydantic 2.
### Description
This pull request added new document loaders to load documents of
various formats using [Dedoc](https://github.com/ispras/dedoc):
- `DedocFileLoader` (determine file types automatically and parse)
- `DedocPDFLoader` (for `PDF` and images parsing)
- `DedocAPIFileLoader` (determine file types automatically and parse
using Dedoc API without library installation)
[Dedoc](https://dedoc.readthedocs.io) is an open-source library/service
that extracts texts, tables, attached files and document structure
(e.g., titles, list items, etc.) from files of various formats. The
library is actively developed and maintained by a group of developers.
`Dedoc` supports `DOCX`, `XLSX`, `PPTX`, `EML`, `HTML`, `PDF`, images
and more.
Full list of supported formats can be found
[here](https://dedoc.readthedocs.io/en/latest/#id1).
For `PDF` documents, `Dedoc` allows to determine textual layer
correctness and split the document into paragraphs.
### Issue
This pull request extends variety of document loaders supported by
`langchain_community` allowing users to choose the most suitable option
for raw documents parsing.
### Dependencies
The PR added a new (optional) dependency `dedoc>=2.2.5` ([library
documentation](https://dedoc.readthedocs.io)) to the
`extended_testing_deps.txt`
### Twitter handle
None
### Add tests and docs
1. Test for the integration:
`libs/community/tests/integration_tests/document_loaders/test_dedoc.py`
2. Example notebook:
`docs/docs/integrations/document_loaders/dedoc.ipynb`
3. Information about the library:
`docs/docs/integrations/providers/dedoc.mdx`
### Lint and test
Done locally:
- `make format`
- `make lint`
- `make integration_tests`
- `make docs_build` (from the project root)
---------
Co-authored-by: Nasty <bogatenkova.anastasiya@mail.ru>
- **Description:** Add a DocumentTransformer for executing one or more
`LinkExtractor`s and adding the extracted links to each document.
- **Issue:** n/a
- **Depedencies:** none
---------
Co-authored-by: Eugene Yurtsev <eugene@langchain.dev>
**Description:** Fixes an issue where the chat message history was not
returned in order. Fixed it now by returning based on timestamps.
- [x] **Add tests and docs**: Updated the tests to check the order
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/docs/integrations` directory.
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
---------
Co-authored-by: Nithish Raghunandanan <nithishr@users.noreply.github.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
This will generate a meaningless string "system: " for generating
condense question; this increases the probability to make an improper
condense question and misunderstand user's question. Below is a case
- Original Question: Can you explain the arguments of Meilisearch?
- Condense Question
- What are the benefits of using Meilisearch? (by CodeLlama)
- What are the reasons for using Meilisearch? (by GPT-4)
The condense questions (not matter from CodeLlam or GPT-4) are different
from the original one.
By checking the content of each dialogue turn, generating history string
only when the dialog content is not empty.
Since there is nothing before first turn, the "history" mechanism will
be ignored at the very first turn.
Doing so, the condense question will be "What are the arguments for
using Meilisearch?".
<!-- 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.
-->
---------
Co-authored-by: Bagatur <baskaryan@gmail.com>
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Co-authored-by: ccurme <chester.curme@gmail.com>
https://www.youtube.com/watch?v=ZIyB9e_7a4c
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
- **Description:**
- Fix#12870: set scope in `default` func (ref:
https://google-auth.readthedocs.io/en/master/reference/google.auth.html)
- Moved the code to load default credentials to the bottom for clarity
of the logic
- Add docstring and comment for each credential loading logic
- **Issue:** https://github.com/langchain-ai/langchain/issues/12870
- **Dependencies:** no dependencies change
- **Tag maintainer:** for a quicker response, tag the relevant
maintainer (see below),
- **Twitter handle:** @gymnstcs
<!-- If no one reviews your PR within a few days, please @-mention one
of @baskaryan, @eyurtsev, @hwchase17.
-->
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- **Description:** `QianfanChatEndpoint` When using tool result to
answer questions, the content of the tool is required to be in Dict
format. Of course, this can require users to return Dict format when
calling the tool, but in order to be consistent with other Chat Models,
I think such modifications are necessary.
- **Description:** Adding notebook to demonstrate visual RAG which uses
both video scene description generated by open source vision models (ex.
video-llama, video-llava etc.) as text embeddings and frames as image
embeddings to perform vector similarity search using VDMS.
- **Issue:** N/A
- **Dependencies:** N/A
Feedback that `RunnableWithMessageHistory` is unwieldy compared to
ConversationChain and similar legacy abstractions is common.
Legacy chains using memory typically had no explicit notion of threads
or separate sessions. To use `RunnableWithMessageHistory`, users are
forced to introduce this concept into their code. This possibly felt
like unnecessary boilerplate.
Here we enable `RunnableWithMessageHistory` to run without a config if
the `get_session_history` callable has no arguments. This enables
minimal implementations like the following:
```python
from langchain_core.chat_history import InMemoryChatMessageHistory
from langchain_core.runnables.history import RunnableWithMessageHistory
from langchain_openai import ChatOpenAI
llm = ChatOpenAI(model="gpt-3.5-turbo-0125")
memory = InMemoryChatMessageHistory()
chain = RunnableWithMessageHistory(llm, lambda: memory)
chain.invoke("Hi I'm Bob") # Hello Bob!
chain.invoke("What is my name?") # Your name is Bob.
```
- **Description:** The correct Prompts for ZERO_SHOT_REACT were not
being used in the `create_sql_agent` function. They were not using the
specific `SQL_PREFIX` and `SQL_SUFFIX` prompts if client does not
provide any prompts. This is fixed.
- **Issue:** #23585
---------
Co-authored-by: ccurme <chester.curme@gmail.com>
Regardless of whether `embedding_func` is set or not, the 'text'
attribute of document should be assigned, otherwise the `page_content`
in the document of the final search result will be lost
### Description
* Fix `libs/langchain/dev.Dockerfile` file. copy the
`libs/standard-tests` folder when building the devcontainer.
* `poetry install --no-interaction --no-ansi --with dev,test,docs`
command requires this folder, but it was not copied.
### Reference
#### Error message when building the devcontainer from the master branch
```
...
[2024-07-20T14:27:34.779Z] ------
> [langchain langchain-dev-dependencies 7/7] RUN poetry install --no-interaction --no-ansi --with dev,test,docs:
0.409
0.409 Directory ../standard-tests does not exist
------
...
```
#### After the fix
Build success at vscode:
<img width="866" alt="image"
src="https://github.com/user-attachments/assets/10db1b50-6fcf-4dfe-83e1-d93c96aa2317">
1. Fix HuggingfacePipeline import error to newer partner package
2. Switch to IPEXModelForCausalLM for performance
There are no dependency changes since optimum intel is also needed for
QuantizedBiEncoderEmbeddings
---------
Co-authored-by: ccurme <chester.curme@gmail.com>
**Description:** Fixes typo `Le'ts` -> `Let's`.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
**Description:**
When initializing retrievers with `configurable_fields` as base
retriever, `ContextualCompressionRetriever` validation fails with the
following error:
```
ValidationError: 1 validation error for ContextualCompressionRetriever
base_retriever
Can't instantiate abstract class BaseRetriever with abstract method _get_relevant_documents (type=type_error)
```
Example code:
```python
esearch_retriever = VertexAISearchRetriever(
project_id=GCP_PROJECT_ID,
location_id="global",
data_store_id=SEARCH_ENGINE_ID,
).configurable_fields(
filter=ConfigurableField(id="vertex_search_filter", name="Vertex Search Filter")
)
# rerank documents with Vertex AI Rank API
reranker = VertexAIRank(
project_id=GCP_PROJECT_ID,
location_id=GCP_REGION,
ranking_config="default_ranking_config",
)
retriever_with_reranker = ContextualCompressionRetriever(
base_compressor=reranker, base_retriever=esearch_retriever
)
```
It seems like the issue stems from ContextualCompressionRetriever
insisting that base retrievers must be strictly `BaseRetriever`
inherited, and doesn't take into account cases where retrievers need to
be chained and can have configurable fields defined.
0a1e475a30/libs/langchain/langchain/retrievers/contextual_compression.py (L15-L22)
This PR proposes that the base_retriever type be set to `RetrieverLike`,
similar to how `EnsembleRetriever` validates its list of retrievers:
0a1e475a30/libs/langchain/langchain/retrievers/ensemble.py (L58-L75)
- **Description:** Add a flag to determine whether to show progress bar
- **Issue:** n/a
- **Dependencies:** n/a
- **Twitter handle:** n/a
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Before, if an exception was raised in the outer `try` block in
`Runnable._atransform_stream_with_config` before `iterator_` is
assigned, the corresponding `finally` block would blow up with an
`UnboundLocalError`:
```txt
UnboundLocalError: cannot access local variable 'iterator_' where it is not associated with a value
```
By assigning an initial value to `iterator_` before entering the `try`
block, this commit ensures that the `finally` can run, and not bury the
"true" exception under a "During handling of the above exception [...]"
traceback.
Thanks for your consideration!
This will allow tools and parsers to accept pydantic models from any of
the
following namespaces:
* pydantic.BaseModel with pydantic 1
* pydantic.BaseModel with pydantic 2
* pydantic.v1.BaseModel with pydantic 2
xfailing some sql tests that do not currently work on sqlalchemy v1
#22207 was very much not sqlalchemy v1 compatible.
Moving forward, implementations should be compatible with both to pass
CI
- **Description:** Search has a limit of 500 results, playlistItems
doesn't. Added a class in except clause to catch another common error.
- **Issue:** None
- **Dependencies:** None
- **Twitter handle:** @TupleType
---------
Co-authored-by: asi-cider <88270351+asi-cider@users.noreply.github.com>
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
**Description:** This PR introduces a change to the
`cypher_generation_chain` to dynamically concatenate inputs. This
improvement aims to streamline the input handling process and make the
method more flexible. The change involves updating the arguments
dictionary with all elements from the `inputs` dictionary, ensuring that
all necessary inputs are dynamically appended. This will ensure that any
cypher generation template will not require a new `_call` method patch.
**Issue:** This PR fixes issue #24260.
The `MongoDBStore` can manage only documents.
It's not possible to use MongoDB for an `CacheBackedEmbeddings`.
With this new implementation, it's possible to use:
```python
CacheBackedEmbeddings.from_bytes_store(
underlying_embeddings=embeddings,
document_embedding_cache=MongoDBByteStore(
connection_string=db_uri,
db_name=db_name,
collection_name=collection_name,
),
)
```
and use MongoDB to cache the embeddings !
- **Description:**
- Updated checksum in doc metadata
- Sending checksum and removing actual content, while sending data to
`pebblo-cloud` if `classifier-location `is `pebblo-cloud` in
`/loader/doc` API
- Adding `pb_id` i.e. pebblo id to doc metadata
- Refactoring as needed.
- Sending `content-checksum` and removing actual content, while sending
data to `pebblo-cloud` if `classifier-location `is `pebblo-cloud` in
`prmopt` API
- **Issue:** NA
- **Dependencies:** NA
- **Tests:** Updated
- **Docs** NA
---------
Co-authored-by: dristy.cd <dristy@clouddefense.io>
Description:
This PR fixes a KeyError: 400 that occurs in the JSON schema processing
within the reduce_openapi_spec function. The _retrieve_ref function in
json_schema.py was modified to handle missing components gracefully by
continuing to the next component if the current one is not found. This
ensures that the OpenAPI specification is fully interpreted and the
agent executes without errors.
Issue:
Fixes issue #24335
Dependencies:
No additional dependencies are required for this change.
Twitter handle:
@lunara_x
**Description:**
**TextEmbed** is a high-performance embedding inference server designed
to provide a high-throughput, low-latency solution for serving
embeddings. It supports various sentence-transformer models and includes
the ability to deploy image and text embedding models. TextEmbed offers
flexibility and scalability for diverse applications.
- **PyPI Package:** [TextEmbed on
PyPI](https://pypi.org/project/textembed/)
- **Docker Image:** [TextEmbed on Docker
Hub](https://hub.docker.com/r/kevaldekivadiya/textembed)
- **GitHub Repository:** [TextEmbed on
GitHub](https://github.com/kevaldekivadiya2415/textembed)
**PR Description**
This PR adds functionality for embedding documents and queries using the
`TextEmbedEmbeddings` class. The implementation allows for both
synchronous and asynchronous embedding requests to a TextEmbed API
endpoint. The class handles batching and permuting of input texts to
optimize the embedding process.
**Example Usage:**
```python
from langchain_community.embeddings import TextEmbedEmbeddings
# Initialise the embeddings class
embeddings = TextEmbedEmbeddings(model="your-model-id", api_key="your-api-key", api_url="your_api_url")
# Define a list of documents
documents = [
"Data science involves extracting insights from data.",
"Artificial intelligence is transforming various industries.",
"Cloud computing provides scalable computing resources over the internet.",
"Big data analytics helps in understanding large datasets.",
"India has a diverse cultural heritage."
]
# Define a query
query = "What is the cultural heritage of India?"
# Embed all documents
document_embeddings = embeddings.embed_documents(documents)
# Embed the query
query_embedding = embeddings.embed_query(query)
# Print embeddings for each document
for i, embedding in enumerate(document_embeddings):
print(f"Document {i+1} Embedding:", embedding)
# Print the query embedding
print("Query Embedding:", query_embedding)
---------
Co-authored-by: Eugene Yurtsev <eugene@langchain.dev>
Fix MultiQueryRetriever breaking Embeddings with empty lines
```
[chain/end] [1:chain:ConversationalRetrievalChain > 2:retriever:Retriever > 3:retriever:Retriever > 4:chain:LLMChain] [2.03s] Exiting Chain run with output:
[outputs]
> /workspaces/Sfeir/sncf/metabot-backend/.venv/lib/python3.11/site-packages/langchain/retrievers/multi_query.py(116)_aget_relevant_documents()
-> if self.include_original:
(Pdb) queries
['## Alternative questions for "Hello, tell me about phones?":', '', '1. **What are the latest trends in smartphone technology?** (Focuses on recent advancements)', '2. **How has the mobile phone industry evolved over the years?** (Historical perspective)', '3. **What are the different types of phones available in the market, and which one is best for me?** (Categorization and recommendation)']
```
Example of failure on VertexAIEmbeddings
```
grpc._channel._InactiveRpcError: <_InactiveRpcError of RPC that terminated with:
status = StatusCode.INVALID_ARGUMENT
details = "The text content is empty."
debug_error_string = "UNKNOWN:Error received from peer ipv4:142.250.184.234:443 {created_time:"2024-04-30T09:57:45.625698408+00:00", grpc_status:3, grpc_message:"The text content is empty."}"
```
Fixes: #15959
---------
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Co-authored-by: Harrison Chase <hw.chase.17@gmail.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- **Description:** Add an async version of `add_documents` to
`ParentDocumentRetriever`
- **Twitter handle:** @johnkdev
---------
Co-authored-by: John Kelly <j.kelly@mwam.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
- **Description:** Add Riza Python/JS code execution tool
- **Issue:** N/A
- **Dependencies:** an optional dependency on the `rizaio` pypi package
- **Twitter handle:** [@rizaio](https://x.com/rizaio)
[Riza](https://riza.io) is a safe code execution environment for
agent-generated Python and JavaScript that's easy to integrate into
langchain apps. This PR adds two new tool classes to the community
package.
- **Description:** Add a `KeybertLinkExtractor` for graph vectorstores.
This allows extracting links from keywords in a Document and linking
nodes that have common keywords.
- **Issue:** None
- **Dependencies:** None.
---------
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
Co-authored-by: ccurme <chester.curme@gmail.com>
- **Description:** This allows extracting links between documents with
common named entities using [GLiNER](https://github.com/urchade/GLiNER).
- **Issue:** None
- **Dependencies:** None
---------
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
This PR updates docs to mention correct version of the
`langchain-openai` package required to use the `stream_usage` parameter.
As it can be noticed in the details of this [merge
commit](722c8f50ea),
that functionality is available only in `langchain-openai >= 0.1.9`
while docs state it's available in `langchain-openai >= 0.1.8`.
- **Description**: Mask API key for ChatOpenAi based chat_models
(openai, azureopenai, anyscale, everlyai).
Made changes to all chat_models that are based on ChatOpenAI since all
of them assumes that openai_api_key is str rather than SecretStr.
- **Issue:**: #12165
- **Dependencies:** N/A
- **Tag maintainer:** @eyurtsev
- **Twitter handle:** N/A
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core,
experimental, etc. is being modified. Use "docs: ..." for purely docs
changes, "templates: ..." for template changes, "infra: ..." for CI
changes.
- Example: "community: add foobar LLM"
- [ ] **PR message**: ***Delete this entire checklist*** and replace
with
- **Description:** a description of the change
- **Issue:** the issue # it fixes, if applicable
- **Dependencies:** any dependencies required for this change
- **Twitter handle:** if your PR gets announced, and you'd like a
mention, we'll gladly shout you out!
- [ ] **Add tests and docs**: 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/docs/integrations` directory.
- [ ] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
Additional guidelines:
- Make sure optional dependencies are imported within a function.
- Please do not add dependencies to pyproject.toml files (even optional
ones) unless they are required for unit tests.
- Most PRs should not touch more than one package.
- Changes should be backwards compatible.
- If you are adding something to community, do not re-import it in
langchain.
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
Description: added support for LangChain v0.2 for nvidia ai endpoint.
Implremented inMemory storage for chains using
RunnableWithMessageHistory which is analogous to using
`ConversationChain` which was used in v0.1 with the default
`ConversationBufferMemory`. This class is deprecated in favor of
`RunnableWithMessageHistory` in LangChain v0.2
Issue: None
Dependencies: None.
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
**Description:**
- Updated the format for the 'Action' section in the planner prompt to
ensure it must be one of the tools without additional words. Adjusted
the phrasing from "should be" to "must be" for clarity and
enforceability.
- Corrected the tool appending logic in the
`_create_api_controller_agent` function to ensure that
`RequestsDeleteToolWithParsing` and `RequestsPatchToolWithParsing` are
properly added to the tools list for "DELETE" and "PATCH" operations.
**Issue:** #24382
**Dependencies:** None
**Twitter handle:** @lunara_x
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Adds MongoDBAtlasVectorSearch to list of VectorStores compatible with
the Indexing API.
(One line change.)
As of `langchain-mongodb = "0.1.7"`, the requirements that the
VectorStore have both add_documents and delete methods with an ids kwarg
is satisfied. #23535 contains the implementation of that, and has been
merged.
Thank you for contributing to LangChain!
- [x] **PR title**: [PebbloSafeLoader] Rename loader type and add
SharePointLoader to supported loaders
- **Description:** Minor fixes in the PebbloSafeLoader:
- Renamed the loader type from `remote_db` to `cloud_folder`.
- Added `SharePointLoader` to the list of loaders supported by
PebbloSafeLoader.
- **Issue:** NA
- **Dependencies:** NA
- [x] **Add tests and docs**: NA
* Please see security warning already in existing class.
* The approach here is fundamentally insecure as it's relying on a block
approach rather than an approach based on only running allowed nodes.
So users should only use this code if its running from a properly
sandboxed environment.
### Description
Missing "stream" parameter. Without it, you'd never receive a stream of
tokens when using stream() or astream()
### Issue
No existing issue available
**Description:** : Add support for chat message history using Couchbase
- [x] **Add tests and docs**: 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/docs/integrations` directory.
- [x] **Lint and test**: Run `make format`, `make lint` and `make test`
from the root of the package(s) you've modified. See contribution
guidelines for more: https://python.langchain.com/docs/contributing/
---------
Co-authored-by: Nithish Raghunandanan <nithishr@users.noreply.github.com>
**Description:**
- Updated constructors in PyPDFParser and PyPDFLoader to handle
`extraction_mode` and additional kwargs, aligning with the capabilities
of `PageObject.extract_text()` from pypdf.
- Added `test_pypdf_loader_with_layout` along with a corresponding
example text file to validate layout extraction from PDFs.
**Issue:** fixes#19735
**Dependencies:** This change requires updating the pypdf dependency
from version 3.4.0 to at least 4.0.0.
Additional changes include the addition of a new test
test_pypdf_loader_with_layout and an example text file to ensure the
functionality of layout extraction from PDFs aligns with the new
capabilities.
---------
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
# Description
This PR aims to solve a bug in `OutputFixingParser`, `RetryOutputParser`
and `RetryWithErrorOutputParser`
The bug is that the wrong keyword argument was given to `retry_chain`.
The correct keyword argument is 'completion', but 'input' is used.
This pull request makes the following changes:
1. correct a `dict` key given to `retry_chain`;
2. add a test when using the default prompt.
- `NAIVE_FIX_PROMPT` for `OutputFixingParser`;
- `NAIVE_RETRY_PROMPT` for `RetryOutputParser`;
- `NAIVE_RETRY_WITH_ERROR_PROMPT` for `RetryWithErrorOutputParser`;
3. ~~add comments on `retry_chain` input and output types~~ clarify
`InputType` and `OutputType` of `retry_chain`
# Issue
The bug is pointed out in
https://github.com/langchain-ai/langchain/pull/19792#issuecomment-2196512928
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
## Description
This pull-request improves the treatment of document IDs in
`MongoDBAtlasVectorSearch`.
Class method signatures of add_documents, add_texts, delete, and
from_texts
now include an `ids:Optional[List[str]]` keyword argument permitting the
user
greater control.
Note that, as before, IDs may also be inferred from
`Document.metadata['_id']`
if present, but this is no longer required,
IDs can also optionally be returned from searches.
This PR closes the following JIRA issues.
* [PYTHON-4446](https://jira.mongodb.org/browse/PYTHON-4446)
MongoDBVectorSearch delete / add_texts function rework
* [PYTHON-4435](https://jira.mongodb.org/browse/PYTHON-4435) Add support
for "Indexing"
* [PYTHON-4534](https://jira.mongodb.org/browse/PYTHON-4534) Ensure
datetimes are json-serializable
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
- Description: When SQLDatabase.from_databricks is ran from a Databricks
Workflow job, line 205 (default_host = context.browserHostName) throws
an ``AttributeError`` as the ``context`` object has no
``browserHostName`` attribute. The fix handles the exception and sets
the ``default_host`` variable to null
---------
Co-authored-by: lmorosdb <lmorosdb>
Co-authored-by: Eugene Yurtsev <eugene@langchain.dev>
**Description:** At the moment neo4j wrapper is using setVectorProperty,
which is deprecated
([link](https://neo4j.com/docs/operations-manual/5/reference/procedures/#procedure_db_create_setVectorProperty)).
I replaced with the non-deprecated version.
Neo4j recently introduced a new cypher method to associate embeddings
into relations using "setRelationshipVectorProperty" method. In this PR
I also implemented a new method to perform this association maintaining
the same format used in the "add_embeddings" method which is used to
associate embeddings into Nodes.
I also included a test case for this new method.
Description: added support for LangChain v0.2 for PipelineAI
integration. Removed deprecated classes and incorporated support for
LangChain v0.2 to integrate with PipelineAI. Removed LLMChain and
replaced it with Runnable interface. Also added StrOutputParser, that
parses LLMResult into the top likely string.
Issue: None
Dependencies: None.
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Alternatively, if you are just interested in using the query generation part of the SQL chain, you can check out [`create_sql_query_chain`](https://github.com/langchain-ai/langchain/blob/master/docs/extras/use_cases/tabular/sql_query.ipynb)
Alternatively, if you are just interested in using the query generation part of the SQL chain, you can check out this [`SQL question-answering tutorial`](https://python.langchain.com/v0.2/docs/tutorials/sql_qa/#convert-question-to-sql-query)
[](https://vscode.dev/redirect?url=vscode://ms-vscode-remote.remote-containers/cloneInVolume?url=https://github.com/langchain-ai/langchain)
[](https://codespaces.new/langchain-ai/langchain)
[llm_symbolic_math.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/llm_symbolic_math.ipynb) | Solve algebraic equations with the help of llms (language learning models) and sympy, a python library for symbolic mathematics.
[meta_prompt.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/meta_prompt.ipynb) | Implement the meta-prompt concept, which is a method for building self-improving agents that reflect on their own performance and modify their instructions accordingly.
[multi_modal_output_agent.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/multi_modal_output_agent.ipynb) | Generate multi-modal outputs, specifically images and text.
[multi_modal_RAG_vdms.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/multi_modal_RAG_vdms.ipynb) | Perform retrieval-augmented generation (rag) on documents including text and images, using unstructured for parsing, Intel's Visual Data Management System (VDMS) as the vectorstore, and chains.
[multi_player_dnd.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/multi_player_dnd.ipynb) | Simulate multi-player dungeons & dragons games, with a custom function determining the speaking schedule of the agents.
[multiagent_authoritarian.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/multiagent_authoritarian.ipynb) | Implement a multi-agent simulation where a privileged agent controls the conversation, including deciding who speaks and when the conversation ends, in the context of a simulated news network.
[multiagent_bidding.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/multiagent_bidding.ipynb) | Implement a multi-agent simulation where agents bid to speak, with the highest bidder speaking next, demonstrated through a fictitious presidential debate example.
@@ -58,4 +59,5 @@ Notebook | Description
[two_player_dnd.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/two_player_dnd.ipynb) | Simulate a two-player dungeons & dragons game, where a dialogue simulator class is used to coordinate the dialogue between the protagonist and the dungeon master.
[wikibase_agent.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/wikibase_agent.ipynb) | Create a simple wikibase agent that utilizes sparql generation, with testing done on http://wikidata.org.
[oracleai_demo.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/oracleai_demo.ipynb) | This guide outlines how to utilize Oracle AI Vector Search alongside Langchain for an end-to-end RAG pipeline, providing step-by-step examples. The process includes loading documents from various sources using OracleDocLoader, summarizing them either within or outside the database with OracleSummary, and generating embeddings similarly through OracleEmbeddings. It also covers chunking documents according to specific requirements using Advanced Oracle Capabilities from OracleTextSplitter, and finally, storing and indexing these documents in a Vector Store for querying with OracleVS.
[rag-locally-on-intel-cpu.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/rag-locally-on-intel-cpu.ipynb) | Perform Retrieval-Augmented-Generation (RAG) on locally downloaded open-source models using langchain and open source tools and execute it on Intel Xeon CPU. We showed an example of how to apply RAG on Llama 2 model and enable it to answer the queries related to Intel Q1 2024 earnings release.
[rag-locally-on-intel-cpu.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/rag-locally-on-intel-cpu.ipynb) | Perform Retrieval-Augmented-Generation (RAG) on locally downloaded open-source models using langchain and open source tools and execute it on Intel Xeon CPU. We showed an example of how to apply RAG on Llama 2 model and enable it to answer the queries related to Intel Q1 2024 earnings release.
[visual_RAG_vdms.ipynb](https://github.com/langchain-ai/langchain/tree/master/cookbook/visual_RAG_vdms.ipynb) | Performs Visual Retrieval-Augmented-Generation (RAG) using videos and scene descriptions generated by open source models.
"This example demonstrates the use of the [SQL Database Agent](/docs/integrations/toolkits/sql_database.html) for answering questions over a Databricks database."
"This example demonstrates the use of the [SQL Database Agent](/docs/integrations/tools/sql_database) for answering questions over a Databricks database."
"* Use of multimodal embeddings (such as [CLIP](https://openai.com/research/clip)) to embed images and text\n",
"* Use of [VDMS](https://github.com/IntelLabs/vdms/blob/master/README.md) as a vector store with support for multi-modal\n",
"* Retrieval of both images and text using similarity search\n",
"* Passing raw images and text chunks to a multimodal LLM for answer synthesis \n",
"\n",
"\n",
"## Packages\n",
"\n",
"For `unstructured`, you will also need `poppler` ([installation instructions](https://pdf2image.readthedocs.io/en/latest/installation.html)) and `tesseract` ([installation instructions](https://tesseract-ocr.github.io/tessdoc/Installation.html)) in your system."
"* Passing raw images and text chunks to a multimodal LLM for answer synthesis "
]
},
{
@@ -53,7 +34,7 @@
},
{
"cell_type": "code",
"execution_count": 3,
"execution_count": 1,
"id": "5f483872",
"metadata": {},
"outputs": [
@@ -61,8 +42,7 @@
"name": "stdout",
"output_type": "stream",
"text": [
"docker: Error response from daemon: Conflict. The container name \"/vdms_rag_nb\" is already in use by container \"0c19ed281463ac10d7efe07eb815643e3e534ddf24844357039453ad2b0c27e8\". You have to remove (or rename) that container to be able to reuse that name.\n",
"For `unstructured`, you will also need `poppler` ([installation instructions](https://pdf2image.readthedocs.io/en/latest/installation.html)) and `tesseract` ([installation instructions](https://tesseract-ocr.github.io/tessdoc/Installation.html)) in your system."
"We can use `partition_pdf` below from [Unstructured](https://unstructured-io.github.io/unstructured/introduction.html#key-concepts) to extract text and images."
"We can use `partition_pdf` from [Unstructured](https://unstructured-io.github.io/unstructured/introduction.html#key-concepts) to extract text and images."
"We will use [OpenClip multimodal embeddings](https://python.langchain.com/docs/integrations/text_embedding/open_clip).\n",
"\n",
"We use a larger model for better performance (set in `langchain_experimental.open_clip.py`).\n",
"\n",
"```\n",
"model_name = \"ViT-g-14\"\n",
"checkpoint = \"laion2b_s34b_b88k\"\n",
"```"
"In this section, we initialize the VDMS vector store for both text and images. For better performance, we use model `ViT-g-14` from [OpenClip multimodal embeddings](https://python.langchain.com/docs/integrations/text_embedding/open_clip).\n",
"The images are stored as base64 encoded strings with `vectorstore.add_images`.\n"
"`vectorstore.add_images` will store / retrieve images as base64 encoded strings."
"Here we define helper functions for image results."
]
},
{
@@ -392,7 +382,8 @@
"id": "1566096d-97c2-4ddc-ba4a-6ef88c525e4e",
"metadata": {},
"source": [
"## Test retrieval and run RAG"
"## Test retrieval and run RAG\n",
"Now let's query for a `woman with children` and retrieve the top results."
]
},
{
@@ -452,6 +443,14 @@
" print(doc.page_content)"
]
},
{
"cell_type": "markdown",
"id": "15e9b54d",
"metadata": {},
"source": [
"Now let's use the `multi_modal_rag_chain` to process the same query and display the response."
]
},
{
"cell_type": "code",
"execution_count": 11,
@@ -462,10 +461,10 @@
"name": "stdout",
"output_type": "stream",
"text": [
"1. Detailed description of the visual elements in the image: The image features a woman with children, likely a mother and her family, standing together outside. They appear to be poor or struggling financially, as indicated by their attire and surroundings.\n",
"2. Historical and cultural context of the image: The photo was taken in 1936 during the Great Depression, when many families struggled to make ends meet. Dorothea Lange, a renowned American photographer, took this iconic photograph that became an emblem of poverty and hardship experienced by many Americans at that time.\n",
"3. Interpretation of the image's symbolism and meaning: The image conveys a sense of unity and resilience despite adversity. The woman and her children are standing together, displaying their strength as a family unit in the face of economic challenges. The photograph also serves as a reminder of the importance of empathy and support for those who are struggling.\n",
"4. Connections between the image and the related text: The text provided offers additional context about the woman in the photo, her background, and her feelings towards the photograph. It highlights the historical backdrop of the Great Depression and emphasizes the significance of this particular image as a representation of that time period.\n"
" The image depicts a woman with several children. The woman appears to be of Cherokee heritage, as suggested by the text provided. The image is described as having been initially regretted by the subject, Florence Owens Thompson, due to her feeling that it did not accurately represent her leadership qualities.\n",
"The historical and cultural context of the image is tied to the Great Depression and the Dust Bowl, both of which affected the Cherokee people in Oklahoma. The photograph was taken during this period, and its subject, Florence Owens Thompson, was a leader within her community who worked tirelessly to help those affected by these crises.\n",
"The image's symbolism and meaning can be interpreted as a representation of resilience and strength in the face of adversity. The woman is depicted with multiple children, which could signify her role as a caregiver and protector during difficult times.\n",
"Connections between the image and the related text include Florence Owens Thompson's leadership qualities and her regretted feelings about the photograph. Additionally, the mention of Dorothea Lange, the photographer who took this photo, ties the image to its historical context and the broader narrative of the Great Depression and Dust Bowl in Oklahoma. \n"
* master color map. Only the colors that actually differ between light and dark
* themes are specified separately.
*
* To see the full list of colors see https://www.figma.com/file/rUrrHGhUBBIAAjQ82x6pz9/PyData-Design-system---proposal-for-implementation-(2)?node-id=1234%3A765&t=ifcFT1JtnrSshGfi-1
..NOTE:: {{objname}} implements the standard :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>`. 🏃
The :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>` has additional methods that are available on runnables, such as :py:meth:`with_types <langchain_core.runnables.base.Runnable.with_types>`, :py:meth:`with_retry <langchain_core.runnables.base.Runnable.with_retry>`, :py:meth:`assign <langchain_core.runnables.base.Runnable.assign>`, :py:meth:`bind <langchain_core.runnables.base.Runnable.bind>`, :py:meth:`get_graph <langchain_core.runnables.base.Runnable.get_graph>`, and more.
..currentmodule:: {{ module }}
..autoclass:: {{ objname }}
..NOTE:: {{objname}} implements the standard :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>`. 🏃
The :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>` has additional methods that are available on runnables, such as :py:meth:`with_types <langchain_core.runnables.base.Runnable.with_types>`, :py:meth:`with_retry <langchain_core.runnables.base.Runnable.with_retry>`, :py:meth:`assign <langchain_core.runnables.base.Runnable.assign>`, :py:meth:`bind <langchain_core.runnables.base.Runnable.bind>`, :py:meth:`get_graph <langchain_core.runnables.base.Runnable.get_graph>`, and more.
..NOTE:: {{objname}} implements the standard :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>`. 🏃
The :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>` has additional methods that are available on runnables, such as :py:meth:`with_types <langchain_core.runnables.base.Runnable.with_types>`, :py:meth:`with_retry <langchain_core.runnables.base.Runnable.with_retry>`, :py:meth:`assign <langchain_core.runnables.base.Runnable.assign>`, :py:meth:`bind <langchain_core.runnables.base.Runnable.bind>`, :py:meth:`get_graph <langchain_core.runnables.base.Runnable.get_graph>`, and more.
..NOTE:: {{objname}} implements the standard :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>`. 🏃
The :py:class:`Runnable Interface <langchain_core.runnables.base.Runnable>` has additional methods that are available on runnables, such as :py:meth:`with_types <langchain_core.runnables.base.Runnable.with_types>`, :py:meth:`with_retry <langchain_core.runnables.base.Runnable.with_retry>`, :py:meth:`assign <langchain_core.runnables.base.Runnable.assign>`, :py:meth:`bind <langchain_core.runnables.base.Runnable.bind>`, :py:meth:`get_graph <langchain_core.runnables.base.Runnable.get_graph>`, and more.
@@ -90,7 +90,7 @@ LCEL aims to provide consistency around behavior and customization over legacy s
`ConversationalRetrievalChain`. Many of these legacy chains hide important details like prompts, and as a wider variety
of viable models emerge, customization has become more and more important.
If you are currently using one of these legacy chains, please see [this guide for guidance on how to migrate](/docs/how_to/migrate_chains/).
If you are currently using one of these legacy chains, please see [this guide for guidance on how to migrate](/docs/versions/migrating_chains).
For guides on how to do specific tasks with LCEL, check out [the relevant how-to guides](/docs/how_to/#langchain-expression-language-lcel).
@@ -165,7 +165,7 @@ Some important things to note:
ChatModels also accept other parameters that are specific to that integration. To find all the parameters supported by a ChatModel head to the API reference for that model.
:::important
**Tool Calling** Some chat models have been fine-tuned for tool calling and provide a dedicated API for tool calling.
Some chat models have been fine-tuned for **tool calling** and provide a dedicated API for it.
Generally, such models are better at tool calling than non-fine-tuned models, and are recommended for use cases that require tool calling.
Please see the [tool calling section](/docs/concepts/#functiontool-calling) for more information.
:::
@@ -236,7 +236,7 @@ This is where information like log-probs and token usage may be stored.
These represent a decision from an language model to call a tool. They are included as part of an `AIMessage` output.
They can be accessed from there with the `.tool_calls` property.
This property returns a list of dictionaries. Each dictionary has the following keys:
This property returns a list of `ToolCall`s. A `ToolCall` is a dictionary with the following arguments:
- `name`: The name of the tool that should be called.
- `args`: The arguments to that tool.
@@ -246,13 +246,18 @@ This property returns a list of dictionaries. Each dictionary has the following
This represents a system message, which tells the model how to behave. Not every model provider supports this.
#### FunctionMessage
This represents the result of a function call. In addition to `role` and `content`, this message has a `name` parameter which conveys the name of the function that was called to produce this result.
#### ToolMessage
This represents the result of a tool call. This is distinct from a FunctionMessage in order to match OpenAI's `function` and `tool` message types. In addition to `role` and `content`, this message has a `tool_call_id` parameter which conveys the id of the call to the tool that was called to produce this result.
This represents the result of a tool call. In addition to `role` and `content`, this message has:
- a `tool_call_id` field which conveys the id of the call to the tool that was called to produce this result.
- an `artifact` field which can be used to pass along arbitrary artifacts of the tool execution which are useful to track but which should not be sent to the model.
#### (Legacy) FunctionMessage
This is a legacy message type, corresponding to OpenAI's legacy function-calling API. `ToolMessage` should be used instead to correspond to the updated tool-calling API.
This represents the result of a function call. In addition to `role` and `content`, this message has a `name` parameter which conveys the name of the function that was called to produce this result.
### Prompt templates
@@ -493,38 +498,130 @@ Retrievers accept a string query as input and return a list of Document's as out
For specifics on how to use retrievers, see the [relevant how-to guides here](/docs/how_to/#retrievers).
### Key-value stores
For some techniques, such as [indexing and retrieval with multiple vectors per document](/docs/how_to/multi_vector/) or
[caching embeddings](/docs/how_to/caching_embeddings/), having a form of key-value (KV) storage is helpful.
LangChain includes a [`BaseStore`](https://api.python.langchain.com/en/latest/stores/langchain_core.stores.BaseStore.html) interface,
which allows for storage of arbitrary data. However, LangChain components that require KV-storage accept a
more specific `BaseStore[str, bytes]` instance that stores binary data (referred to as a `ByteStore`), and internally take care of
encoding and decoding data for their specific needs.
This means that as a user, you only need to think about one type of store rather than different ones for different types of data.
#### Interface
All [`BaseStores`](https://api.python.langchain.com/en/latest/stores/langchain_core.stores.BaseStore.html) support the following interface. Note that the interface allows
for modifying **multiple** key-value pairs at once:
- `mget(key: Sequence[str]) -> List[Optional[bytes]]`: get the contents of multiple keys, returning `None` if the key does not exist
- `mset(key_value_pairs: Sequence[Tuple[str, bytes]]) -> None`: set the contents of multiple keys
- `yield_keys(prefix: Optional[str] = None) -> Iterator[str]`: yield all keys in the store, optionally filtering by a prefix
For key-value store implementations, see [this section](/docs/integrations/stores/).
### Tools
<span data-heading-keywords="tool,tools"></span>
Tools are interfaces that an agent, a chain, or a chat model / LLM can use to interact with the world.
Tools are utilities designed to be called by a model: their inputs are designed to be generated by models, and their outputs are designed to be passed back to models.
Tools are needed whenever you want a model to control parts of your code or call out to external APIs.
A tool consists of the following components:
A tool consists of:
1. The name of the tool
2. A description of what the tool does
3. JSON schema of what the inputs to the tool are
4. The function to call
5. Whether the result of a tool should be returned directly to the user (only relevant for agents)
1. The name of the tool.
2. A description of what the tool does.
3. A JSON schema defining the inputs to the tool.
4. A function (and, optionally, an async variant of the function).
The name, description and JSON schema are provided as context
to the LLM, allowing the LLM to determine how to use the tool
appropriately.
When a tool is bound to a model, the name, description and JSON schema are provided as context to the model.
Given a list of tools and a set of instructions, a model can request to call one or more tools with specific inputs.
Typical usage may look like the following:
Given a list of available tools and a prompt, an LLM can request
that one or more tools be invoked with appropriate arguments.
Generally, when designing tools to be used by a chat model or LLM, it is important to keep in mind the following:
The `AIMessage` returned from the model MAY have `tool_calls` associated with it.
Read [this guide](/docs/concepts/#aimessage) for more information on what the response type may look like.
- Chat models that have been fine-tuned for tool calling will be better at tool calling than non-fine-tuned models.
- Non fine-tuned models may not be able to use tools at all, especially if the tools are complex or require multiple tool calls.
- Models will perform better if the tools have well-chosen names, descriptions, and JSON schemas.
- Simpler tools are generally easier for models to use than more complex tools.
Once the chosen tools are invoked, the results can be passed back to the model so that it can complete whatever task
it's performing.
There are generally two different ways to invoke the tool and pass back the response:
For specifics on how to use tools, see the [relevant how-to guides here](/docs/how_to/#tools).
#### Invoke with just the arguments
To use an existing pre-built tool, see [here](/docs/integrations/tools/) for a list of pre-built tools.
When you invoke a tool with just the arguments, you will get back the raw tool output (usually a string).
This generally looks like:
```python
# You will want to previously check that the LLM returned tool calls
tool_call = ai_msg.tool_calls[0]
# ToolCall(args={...}, id=..., ...)
tool_output = tool.invoke(tool_call["args"])
tool_message = ToolMessage(
content=tool_output,
tool_call_id=tool_call["id"],
name=tool_call["name"]
)
```
Note that the `content` field will generally be passed back to the model.
If you do not want the raw tool response to be passed to the model, but you still want to keep it around,
you can transform the tool output but also pass it as an artifact (read more about [`ToolMessage.artifact` here](/docs/concepts/#toolmessage))
```python
... # Same code as above
response_for_llm = transform(response)
tool_message = ToolMessage(
content=response_for_llm,
tool_call_id=tool_call["id"],
name=tool_call["name"],
artifact=tool_output
)
```
#### Invoke with `ToolCall`
The other way to invoke a tool is to call it with the full `ToolCall` that was generated by the model.
When you do this, the tool will return a ToolMessage.
The benefits of this are that you don't have to write the logic yourself to transform the tool output into a ToolMessage.
This generally looks like:
```python
tool_call = ai_msg.tool_calls[0]
# -> ToolCall(args={...}, id=..., ...)
tool_message = tool.invoke(tool_call)
# -> ToolMessage(
content="tool result foobar...",
tool_call_id=...,
name="tool_name"
)
```
If you are invoking the tool this way and want to include an [artifact](/docs/concepts/#toolmessage) for the ToolMessage, you will need to have the tool return two things.
Read more about [defining tools that return artifacts here](/docs/how_to/tool_artifacts/).
#### Best practices
When designing tools to be used by a model, it is important to keep in mind that:
- Chat models that have explicit [tool-calling APIs](/docs/concepts/#functiontool-calling) will be better at tool calling than non-fine-tuned models.
- Models will perform better if the tools have well-chosen names, descriptions, and JSON schemas. This another form of prompt engineering.
- Simple, narrowly scoped tools are easier for models to use than complex tools.
#### Related
For specifics on how to use tools, see the [tools how-to guides](/docs/how_to/#tools).
To use a pre-built tool, see the [tool integration docs](/docs/integrations/tools/).
Toolkits are collections of tools that are designed to be used together for specific tasks. They have convenient loading methods.
@@ -769,6 +866,61 @@ units (like words or subwords) that carry meaning, rather than individual charac
to learn and understand the structure of the language, including grammar and context.
Furthermore, using tokens can also improve efficiency, since the model processes fewer units of text compared to character-level processing.
### Function/tool calling
:::info
We use the term tool calling interchangeably with function calling. Although
function calling is sometimes meant to refer to invocations of a single function,
we treat all models as though they can return multiple tool or function calls in
each message.
:::
Tool calling allows a [chat model](/docs/concepts/#chat-models) to respond to a given prompt by generating output that
matches a user-defined schema.
While the name implies that the model is performing
some action, this is actually not the case! The model only generates the arguments to a tool, and actually running the tool (or not) is up to the user.
One common example where you **wouldn't** want to call a function with the generated arguments
is if you want to [extract structured output matching some schema](/docs/concepts/#structured-output)
from unstructured text. You would give the model an "extraction" tool that takes
parameters matching the desired schema, then treat the generated output as your final
result.
Tool calling is not universal, but is supported by many popular LLM providers, including [Anthropic](/docs/integrations/chat/anthropic/),
[Mistral](/docs/integrations/chat/mistralai/), [OpenAI](/docs/integrations/chat/openai/), and even for locally-running models via [Ollama](/docs/integrations/chat/ollama/).
LangChain provides a standardized interface for tool calling that is consistent across different models.
The standard interface consists of:
* `ChatModel.bind_tools()`: a method for specifying which tools are available for a model to call. This method accepts [LangChain tools](/docs/concepts/#tools) as well as [Pydantic](https://pydantic.dev/) objects.
* `AIMessage.tool_calls`: an attribute on the `AIMessage` returned from the model for accessing the tool calls requested by the model.
#### Tool usage
After the model calls tools, you can use the tool by invoking it, then passing the arguments back to the model.
LangChain provides the [`Tool`](/docs/concepts/#tools) abstraction to help you handle this.
The general flow is this:
1. Generate tool calls with a chat model in response to a query.
2. Invoke the appropriate tools using the generated tool call as arguments.
3. Format the result of the tool invocations as [`ToolMessages`](/docs/concepts/#toolmessage).
4. Pass the entire list of messages back to the model so that it can generate a final answer (or call more tools).
This is how tool calling [agents](/docs/concepts/#agents) perform tasks and answer queries.
Check out some more focused guides below:
- [How to use chat models to call tools](/docs/how_to/tool_calling/)
- [How to pass tool outputs to chat models](/docs/how_to/tool_results_pass_to_model/)
- [Building an agent with LangGraph](https://langchain-ai.github.io/langgraph/tutorials/introduction/)
### Structured output
LLMs are capable of generating arbitrary text. This enables the model to respond appropriately to a wide
@@ -901,48 +1053,48 @@ chain.invoke({ "question": "What is the powerhouse of the cell?" })
For a full list of model providers that support JSON mode, see [this table](/docs/integrations/chat/#advanced-features).
[Mistral](https://mistral.ai/), [OpenAI](https://openai.com/), and others,
support variants of a tool calling feature. These features typically allow requests
to the LLM to include available tools and their schemas, and for responses to include
calls to these tools. For instance, given a search engine tool, an LLM might handle a
query by first issuing a call to the search engine. The system calling the LLM can
receive the tool call, execute it, and return the output to the LLM to inform its
response. LangChain includes a suite of [built-in tools](/docs/integrations/tools/)
and supports several methods for defining your own [custom tools](/docs/how_to/custom_tools).
```python
from langchain_core.pydantic_v1 import BaseModel, Field
from langchain_openai import ChatOpenAI
LangChain provides a standardized interface for tool calling that is consistent across different models.
class ResponseFormatter(BaseModel):
"""Always use this tool to structure your response to the user."""
The standard interface consists of:
answer: str = Field(description="The answer to the user's question")
followup_question: str = Field(description="A followup question the user could ask")
* `ChatModel.bind_tools()`: a method for specifying which tools are available for a model to call. This method accepts [LangChain tools](/docs/concepts/#tools) here.
* `AIMessage.tool_calls`: an attribute on the `AIMessage` returned from the model for accessing the tool calls requested by the model.
model = ChatOpenAI(
model="gpt-4o",
temperature=0,
)
The following how-to guides are good practical resources for using function/tool calling:
ai_msg = model_with_tools.invoke("What is the powerhouse of the cell?")
ai_msg.tool_calls[0]["args"]
```
```
{'answer': "The powerhouse of the cell is the mitochondrion. It generates most of the cell's supply of adenosine triphosphate (ATP), which is used as a source of chemical energy.",
'followup_question': 'How do mitochondria generate ATP?'}
```
Tool calling is a generally consistent way to get a model to generate structured output, and is the default technique
used for the [`.with_structured_output()`](/docs/concepts/#with_structured_output) method when a model supports it.
The following how-to guides are good practical resources for using function/tool calling for structured output:
- [How to return structured data from an LLM](/docs/how_to/structured_output/)
- [How to use a model to call tools](/docs/how_to/tool_calling)
"This functionality was added in ``langchain-core == 0.2.24``. Please make sure your package is up to date.\n",
":::"
]
},
{
"cell_type": "markdown",
"id": "cbc3c873-6109-4e03-b775-b73c1003faea",
"metadata": {},
"source": [
"## Initialize a rate limiter\n",
"\n",
"Langchain comes with a built-in in memory rate limiter. This rate limiter is thread safe and can be shared by multiple threads in the same process.\n",
"\n",
"The provided rate limiter can only limit the number of requests per unit time. It will not help if you need to also limited based on the size\n",
"Tracking token usage to calculate cost is an important part of putting your app in production. This guide goes over how to obtain this information from your LangChain model calls.\n",
"For example, OpenAI will return a message [chunk](https://api.python.langchain.com/en/latest/messages/langchain_core.messages.ai.AIMessageChunk.html) at the end of a stream with token usage information. This behavior is supported by `langchain-openai >= 0.1.8` and can be enabled by setting `stream_usage=True`. This attribute can also be set when `ChatOpenAI` is instantiated.\n",
"For example, OpenAI will return a message [chunk](https://api.python.langchain.com/en/latest/messages/langchain_core.messages.ai.AIMessageChunk.html) at the end of a stream with token usage information. This behavior is supported by `langchain-openai >= 0.1.9` and can be enabled by setting `stream_usage=True`. This attribute can also be set when `ChatOpenAI` is instantiated.\n",
"[LLMListwiseRerank](https://api.python.langchain.com/en/latest/retrievers/langchain.retrievers.document_compressors.listwise_rerank.LLMListwiseRerank.html) uses [zero-shot listwise document reranking](https://arxiv.org/pdf/2305.02156) and functions similarly to `LLMChainFilter` as a robust but more expensive option. It is recommended to use a more powerful LLM.\n",
"\n",
"Note that `LLMListwiseRerank` requires a model with the [with_structured_output](/docs/integrations/chat/) method implemented."
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "4ab9ee9f-917e-4d6f-9344-eb7f01533228",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"Document 1:\n",
"\n",
"Tonight. I call on the Senate to: Pass the Freedom to Vote Act. Pass the John Lewis Voting Rights Act. And while you’re at it, pass the Disclose Act so Americans can know who is funding our elections. \n",
"\n",
"Tonight, I’d like to honor someone who has dedicated his life to serve this country: Justice Stephen Breyer—an Army veteran, Constitutional scholar, and retiring Justice of the United States Supreme Court. Justice Breyer, thank you for your service. \n",
"\n",
"One of the most serious constitutional responsibilities a President has is nominating someone to serve on the United States Supreme Court. \n",
"\n",
"And I did that 4 days ago, when I nominated Circuit Court of Appeals Judge Ketanji Brown Jackson. One of our nation’s top legal minds, who will continue Justice Breyer’s legacy of excellence.\n"
"When constructing an agent, you will need to provide it with a list of `Tool`s that it can use. Besides the actual function that is called, the Tool consists of several components:\n",
"\n",
@@ -270,7 +270,7 @@
"source": [
"### StructuredTool\n",
"\n",
"The `StrurcturedTool.from_function` class method provides a bit more configurability than the `@tool` decorator, without requiring much additional code."
"The `StructuredTool.from_function` class method provides a bit more configurability than the `@tool` decorator, without requiring much additional code."
"* The `load` methods is a convenience method meant solely for prototyping work -- it just invokes `list(self.lazy_load())`.\n",
"* The `alazy_load` has a default implementation that will delegate to `lazy_load`. If you're using async, we recommend overriding the default implementation and providing a native async implementation.\n",
"\n",
":::{.callout-important}\n",
":::{.callout-important}\n",
"When implementing a document loader do **NOT** provide parameters via the `lazy_load` or `alazy_load` methods.\n",
"\n",
"All configuration is expected to be passed through the initializer (__init__). This was a design choice made by LangChain to make sure that once a document loader has been instantiated it has all the information needed to load documents.\n",
@@ -235,7 +235,7 @@
"id": "56cb443e-f987-4386-b4ec-975ee129adb2",
"metadata": {},
"source": [
":::{.callout-tip}\n",
":::{.callout-tip}\n",
"\n",
"`load()` can be helpful in an interactive environment such as a jupyter notebook.\n",
"\n",
@@ -276,7 +276,7 @@
"source": [
"## Working with Files\n",
"\n",
"Many document loaders invovle parsing files. The difference between such loaders usually stems from how the file is parsed rather than how the file is loaded. For example, you can use `open` to read the binary content of either a PDF or a markdown file, but you need different parsing logic to convert that binary data into text.\n",
"Many document loaders involve parsing files. The difference between such loaders usually stems from how the file is parsed, rather than how the file is loaded. For example, you can use `open` to read the binary content of either a PDF or a markdown file, but you need different parsing logic to convert that binary data into text.\n",
"\n",
"As a result, it can be helpful to decouple the parsing logic from the loading logic, which makes it easier to re-use a given parser regardless of how the data was loaded.\n",
"\n",
@@ -355,7 +355,7 @@
"id": "433bfb7c-7767-43bc-b71e-42413d7494a8",
"metadata": {},
"source": [
"Using the **blob** API also allows one to load content direclty from memory without having to read it from a file!"
"Using the **blob** API also allows one to load content directly from memory without having to read it from a file!"
"You can use arbitrary functions as [Runnables](https://api.python.langchain.com/en/latest/runnables/langchain_core.runnables.base.Runnable.html#langchain_core.runnables.base.Runnable). This is useful for formatting or when you need functionality not provided by other LangChain components, and custom functions used as Runnables are called [`RunnableLambdas`](https://api.python.langchain.com/en/latest/runnables/langchain_core.runnables.base.RunnableLambda.html).\n",
"\n",
"Note that all inputs to these functions need to be a SINGLE argument. If you have a function that accepts multiple arguments, you should write a wrapper that accepts a single dict input and unpacks it into multiple argument.\n",
"Note that all inputs to these functions need to be a SINGLE argument. If you have a function that accepts multiple arguments, you should write a wrapper that accepts a single dict input and unpacks it into multiple arguments.\n",
@@ -31,6 +31,8 @@ This highlights functionality that is core to using LangChain.
[**LCEL cheatsheet**](/docs/how_to/lcel_cheatsheet/): For a quick overview of how to use the main LCEL primitives.
[**Migration guide**](/docs/versions/migrating_chains): For migrating legacy chain abstractions to LCEL.
- [How to: chain runnables](/docs/how_to/sequence)
- [How to: stream runnables](/docs/how_to/streaming)
- [How to: invoke runnables in parallel](/docs/how_to/parallel/)
@@ -43,7 +45,7 @@ This highlights functionality that is core to using LangChain.
- [How to: create a dynamic (self-constructing) chain](/docs/how_to/dynamic_chain/)
- [How to: inspect runnables](/docs/how_to/inspect)
- [How to: add fallbacks to a runnable](/docs/how_to/fallbacks)
- [How to: migrate chains to LCEL](/docs/how_to/migrate_chains)
- [How to: pass runtime secrets to a runnable](/docs/how_to/runnable_runtime_secrets)
## Components
@@ -80,12 +82,13 @@ These are the core building blocks you can use when building applications.
- [How to: stream a response back](/docs/how_to/chat_streaming)
- [How to: track token usage](/docs/how_to/chat_token_usage_tracking)
- [How to: track response metadata across providers](/docs/how_to/response_metadata)
- [How to: let your end users choose their model](/docs/how_to/chat_models_universal_init/)
- [How to: use chat model to call tools](/docs/how_to/tool_calling)
- [How to: stream tool calls](/docs/how_to/tool_streaming)
- [How to: handle rate limits](/docs/how_to/chat_model_rate_limiting)
- [How to: few shot prompt tool behavior](/docs/how_to/tools_few_shot)
- [How to: bind model-specific formatted tools](/docs/how_to/tools_model_specific)
- [How to: force a specific tool call](/docs/how_to/tool_choice)
- [How to: work with local models](/docs/how_to/local_llms)
- [How to: init any model in one line](/docs/how_to/chat_models_universal_init/)
### Messages
@@ -104,7 +107,7 @@ What LangChain calls [LLMs](/docs/concepts/#llms) are older forms of language mo
- [How to: create a custom LLM class](/docs/how_to/custom_llm)
- [How to: stream a response back](/docs/how_to/streaming_llm)
- [How to: track token usage](/docs/how_to/llm_token_usage_tracking)
- [How to: work with local LLMs](/docs/how_to/local_llms)
- [How to: work with local models](/docs/how_to/local_llms)
### Output parsers
@@ -185,19 +188,21 @@ Indexing is the process of keeping your vectorstore in-sync with the underlying
LangChain [Tools](/docs/concepts/#tools) contain a description of the tool (to pass to the language model) as well as the implementation of the function to call. Refer [here](/docs/integrations/tools/) for a list of pre-buit tools.
- [How to: create custom tools](/docs/how_to/custom_tools)
- [How to: use built-in tools and built-in toolkits](/docs/how_to/tools_builtin)
- [How to: convert Runnables to tools](/docs/how_to/convert_runnable_to_tool)
- [How to: use chat model to call tools](/docs/how_to/tool_calling)
- [How to: pass tool results back to model](/docs/how_to/tool_results_pass_to_model)
- [How to: add ad-hoc tool calling capability to LLMs and chat models](/docs/how_to/tools_prompting)
- [How to: create tools](/docs/how_to/custom_tools)
- [How to: use built-in tools and toolkits](/docs/how_to/tools_builtin)
- [How to: use chat models to call tools](/docs/how_to/tool_calling)
- [How to: pass tool outputs to chat models](/docs/how_to/tool_results_pass_to_model)
- [How to: pass run time values to tools](/docs/how_to/tool_runtime)
- [How to: add a humanintheloop to tool usage](/docs/how_to/tools_human)
- [How to: handle errors when calling tools](/docs/how_to/tools_error)
- [How to: disable parallel tool calling](/docs/how_to/tool_choice)
- [How to: access the `RunnableConfig` object within a custom tool](/docs/how_to/tool_configure)
- [How to: stream events from child runs within a custom tool](/docs/how_to/tool_stream_events)
- [How to: return extra artifacts from a tool](/docs/how_to/tool_artifacts/)
- [How to: add a human-in-the-loop for tools](/docs/how_to/tools_human)
- [How to: handle tool errors](/docs/how_to/tools_error)
- [How to: force models to call a tool](/docs/how_to/tool_choice)
- [How to: disable parallel tool calling](/docs/how_to/tool_calling_parallel)
- [How to: access the `RunnableConfig` from a tool](/docs/how_to/tool_configure)
- [How to: stream events from a tool](/docs/how_to/tool_stream_events)
- [How to: return artifacts from a tool](/docs/how_to/tool_artifacts/)
- [How to: convert Runnables to tools](/docs/how_to/convert_runnable_to_tool)
- [How to: add ad-hoc tool calling capability to models](/docs/how_to/tools_prompting)
- [How to: pass in runtime secrets](/docs/how_to/runnable_runtime_secrets)
"The popularity of projects like [PrivateGPT](https://github.com/imartinez/privateGPT), [llama.cpp](https://github.com/ggerganov/llama.cpp), [Ollama](https://github.com/ollama/ollama), [GPT4All](https://github.com/nomic-ai/gpt4all), [llamafile](https://github.com/Mozilla-Ocho/llamafile), and others underscore the demand to run LLMs locally (on your own device).\n",
"The popularity of projects like [llama.cpp](https://github.com/ggerganov/llama.cpp), [Ollama](https://github.com/ollama/ollama), [GPT4All](https://github.com/nomic-ai/gpt4all), [llamafile](https://github.com/Mozilla-Ocho/llamafile), and others underscore the demand to run LLMs locally (on your own device).\n",
"Some providers have [chat model](/docs/concepts/#chat-models) wrappers that takes care of formatting your input prompt for the specific local model you're using. However, if you are prompting local models with a [text-in/text-out LLM](/docs/concepts/#llms) wrapper, you may need to use a prompt tailed for your specific model.\n",
"\n",
"This can [require the inclusion of special tokens](https://huggingface.co/blog/llama2#how-to-prompt-llama-2). [Here's an example for LLaMA 2](https://smith.langchain.com/hub/rlm/rag-prompt-llama).\n",
"\n",
"## Quickstart\n",
"\n",
"[`Ollama`](https://ollama.ai/) is one way to easily run inference on macOS.\n",
@@ -73,10 +79,20 @@
"The instructions [here](https://github.com/jmorganca/ollama?tab=readme-ov-file#ollama) provide details, which we summarize:\n",
" \n",
"* [Download and run](https://ollama.ai/download) the app\n",
"* From command line, fetch a model from this [list of options](https://github.com/jmorganca/ollama): e.g., `ollama pull llama2`\n",
"* From command line, fetch a model from this [list of options](https://github.com/jmorganca/ollama): e.g., `ollama pull llama3.1:8b`\n",
"* When the app is running, all models are automatically served on `localhost:11434`\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"id": "29450fc9",
"metadata": {},
"outputs": [],
"source": [
"%pip install -qU langchain_ollama"
]
},
{
"cell_type": "code",
"execution_count": 2,
@@ -86,7 +102,7 @@
{
"data": {
"text/plain": [
"' The first man on the moon was Neil Armstrong, who landed on the moon on July 20, 1969 as part of the Apollo 11 mission. obviously.'"
"'...Neil Armstrong!\\n\\nOn July 20, 1969, Neil Armstrong became the first person to set foot on the lunar surface, famously declaring \"That\\'s one small step for man, one giant leap for mankind\" as he stepped off the lunar module Eagle onto the Moon\\'s surface.\\n\\nWould you like to know more about the Apollo 11 mission or Neil Armstrong\\'s achievements?'"
]
},
"execution_count": 2,
@@ -95,51 +111,78 @@
}
],
"source": [
"from langchain_community.llms import Ollama\n",
"from langchain_ollama import OllamaLLM\n",
"\n",
"llm = OllamaLLM(model=\"llama3.1:8b\")\n",
"\n",
"llm = Ollama(model=\"llama2\")\n",
"llm.invoke(\"The first man on the moon was ...\")"
]
},
{
"cell_type": "markdown",
"id": "343ab645",
"id": "674cc672",
"metadata": {},
"source": [
"Stream tokens as they are being generated."
"Stream tokens as they are being generated:"
]
},
{
"cell_type": "code",
"execution_count": 40,
"id": "9cd83603",
"execution_count": 3,
"id": "1386a852",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
" The first man to walk on the moon was Neil Armstrong, an American astronaut who was part of the Apollo 11 mission in 1969. февруари 20, 1969, Armstrong stepped out of the lunar module Eagle and onto the moon's surface, famously declaring \"That's one small step for man, one giant leap for mankind\" as he took his first steps. He was followed by fellow astronaut Edwin \"Buzz\" Aldrin, who also walked on the moon during the mission."
"for chunk in llm.stream(\"The first man on the moon was ...\"):\n",
" print(chunk, end=\"|\", flush=True)"
]
},
{
"cell_type": "markdown",
"id": "e5731060",
"metadata": {},
"source": [
"Ollama also includes a chat model wrapper that handles formatting conversation turns:"
]
},
{
"cell_type": "code",
"execution_count": 4,
"id": "f14a778a",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"' The first man to walk on the moon was Neil Armstrong, an American astronaut who was part of the Apollo 11 mission in 1969. февруари 20, 1969, Armstrong stepped out of the lunar module Eagle and onto the moon\\'s surface, famously declaring\"That\\'s one small step for man, one giant leap for mankind\" as he took his first steps. He was followed by fellow astronaut Edwin \"Buzz\" Aldrin, who also walked on the moon during the mission.'"
"AIMessage(content='The answer is a historic one!\\n\\nThe first man to walk on the Moon was Neil Armstrong, an American astronaut and commander of the Apollo 11 mission. On July 20, 1969, Armstrong stepped out of the lunar module Eagle onto the surface of the Moon, famously declaring:\\n\\n\"That\\'s one small step for man, one giant leap for mankind.\"\\n\\nArmstrong was followed by fellow astronaut Edwin \"Buzz\" Aldrin, who also walked on the Moon during the mission. Michael Collins remained in orbit around the Moon in the command module Columbia.\\n\\nNeil Armstrong passed away on August 25, 2012, but his legacy as a pioneering astronaut and engineer continues to inspire people around the world!', response_metadata={'model': 'llama3.1:8b', 'created_at': '2024-08-01T00:38:29.176717Z', 'message': {'role': 'assistant', 'content': ''}, 'done_reason': 'stop', 'done': True, 'total_duration': 10681861417, 'load_duration': 34270292, 'prompt_eval_count': 19, 'prompt_eval_duration': 6209448000, 'eval_count': 141, 'eval_duration': 4432022000}, id='run-7bed57c5-7f54-4092-912c-ae49073dcd48-0', usage_metadata={'input_tokens': 19, 'output_tokens': 141, 'total_tokens': 160})"
"chat_model.invoke(\"Who was the first man on the moon?\")"
]
},
{
@@ -199,7 +242,7 @@
"\n",
"With [Ollama](https://github.com/jmorganca/ollama), fetch a model via `ollama pull <model family>:<tag>`:\n",
"\n",
"* E.g., for Llama-7b: `ollama pull llama2` will download the most basic version of the model (e.g., smallest # parameters and 4 bit quantization)\n",
"* E.g., for Llama 2 7b: `ollama pull llama2` will download the most basic version of the model (e.g., smallest # parameters and 4 bit quantization)\n",
"* We can also specify a particular version from the [model list](https://github.com/jmorganca/ollama?tab=readme-ov-file#model-library), e.g., `ollama pull llama2:13b`\n",
"* See the full set of parameters on the [API reference page](https://api.python.langchain.com/en/latest/llms/langchain_community.llms.ollama.Ollama.html)"
]
@@ -222,9 +265,7 @@
}
],
"source": [
"from langchain_community.llms import Ollama\n",
"\n",
"llm = Ollama(model=\"llama2:13b\")\n",
"llm = OllamaLLM(model=\"llama2:13b\")\n",
"llm.invoke(\"The first man on the moon was ... think step by step\")"
" 'output': 'The value of `magic_function(3)` is 5.'}"
]
},
"execution_count": 2,
"execution_count": 3,
"metadata": {},
"output_type": "execute_result"
}
@@ -142,7 +162,7 @@
},
{
"cell_type": "code",
"execution_count": 3,
"execution_count": 4,
"id": "53a3737a-d167-4255-89bf-20ac37f89a3e",
"metadata": {},
"outputs": [
@@ -153,7 +173,7 @@
" 'output': 'The value of `magic_function(3)` is 5.'}"
]
},
"execution_count": 3,
"execution_count": 4,
"metadata": {},
"output_type": "execute_result"
}
@@ -173,7 +193,7 @@
},
{
"cell_type": "code",
"execution_count": 4,
"execution_count": 5,
"id": "74ecebe3-512e-409c-a661-bdd5b0a2b782",
"metadata": {},
"outputs": [
@@ -181,10 +201,10 @@
"data": {
"text/plain": [
"{'input': 'Pardon?',\n",
" 'output': 'The result of applying `magic_function` to the input 3 is 5.'}"
" 'output': 'The value you get when you apply `magic_function` to the input 3 is 5.'}"
]
},
"execution_count": 4,
"execution_count": 5,
"metadata": {},
"output_type": "execute_result"
}
@@ -223,7 +243,7 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 6,
"id": "a9a11ccd-75e2-4c11-844d-a34870b0ff91",
"metadata": {},
"outputs": [
@@ -234,7 +254,7 @@
" 'output': 'El valor de `magic_function(3)` es 5.'}"
]
},
"execution_count": 5,
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
@@ -263,19 +283,19 @@
"source": [
"Now, let's pass a custom system message to [react agent executor](https://langchain-ai.github.io/langgraph/reference/prebuilt/#create_react_agent).\n",
"\n",
"LangGraph's prebuilt `create_react_agent` does not take a prompt template directly as a parameter, but instead takes a [`messages_modifier`](https://langchain-ai.github.io/langgraph/reference/prebuilt/#create_react_agent) parameter. This modifies messages before they are passed into the model, and can be one of four values:\n",
"LangGraph's prebuilt `create_react_agent` does not take a prompt template directly as a parameter, but instead takes a [`state_modifier`](https://langchain-ai.github.io/langgraph/reference/prebuilt/#create_react_agent) parameter. This modifies the graph state before the llm is called, and can be one of four values:\n",
"\n",
"- A `SystemMessage`, which is added to the beginning of the list of messages.\n",
"- A `string`, which is converted to a `SystemMessage` and added to the beginning of the list of messages.\n",
"- A `Callable`, which should take in a list of messages. The output is then passed to the language model.\n",
"- Or a [`Runnable`](/docs/concepts/#langchain-expression-language-lcel), which should should take in a list of messages. The output is then passed to the language model.\n",
"- A `Callable`, which should take in full graph state. The output is then passed to the language model.\n",
"- Or a [`Runnable`](/docs/concepts/#langchain-expression-language-lcel), which should take in full graph state. The output is then passed to the language model.\n",
"\n",
"Here's how it looks in action:"
]
},
{
"cell_type": "code",
"execution_count": 6,
"execution_count": 7,
"id": "a9486805-676a-4d19-a5c4-08b41b172989",
"metadata": {},
"outputs": [],
@@ -287,7 +307,7 @@
"# This could also be a SystemMessage object\n",
"# system_message = SystemMessage(content=\"You are a helpful assistant. Respond only in Spanish.\")\n",
" AIMessage(content='The value of `magic_function(3)` is 5.', response_metadata={'token_usage': {'completion_tokens': 14, 'prompt_tokens': 78, 'total_tokens': 92}, 'model_name': 'gpt-4o-2024-05-13', 'system_fingerprint': 'fp_4e2b2da518', 'finish_reason': 'stop', 'logprobs': None}, id='run-abf9341c-ef41-4157-935d-a3be5dfa2f41-0', usage_metadata={'input_tokens': 78, 'output_tokens': 14, 'total_tokens': 92})]}"
]
},
"execution_count": 13,
@@ -708,7 +726,7 @@
},
{
"cell_type": "code",
"execution_count": 14,
"execution_count": 16,
"id": "16f189a7-fc78-4cb5-aa16-a94ca06401a6",
"metadata": {},
"outputs": [],
@@ -724,7 +742,7 @@
},
{
"cell_type": "code",
"execution_count": 15,
"execution_count": 17,
"id": "c96aefd7-6f6e-4670-aca6-1ac3d4e7871f",
"metadata": {},
"outputs": [
@@ -739,11 +757,7 @@
"Invoking: `magic_function` with `{'input': '3'}`\n",
"\n",
"\n",
"\u001b[0m\u001b[36;1m\u001b[1;3mSorry, there was an error. Please try again.\u001b[0m\u001b[32;1m\u001b[1;3m\n",
"Invoking: `magic_function` with `{'input': '3'}`\n",
"responded: Parece que hubo un error al intentar obtener el valor de `magic_function(3)`. Permíteme intentarlo de nuevo.\n",
"\n",
"\u001b[0m\u001b[36;1m\u001b[1;3mSorry, there was an error. Please try again.\u001b[0m\u001b[32;1m\u001b[1;3mAún no puedo obtener el valor de `magic_function(3)`. ¿Hay algo más en lo que pueda ayudarte?\u001b[0m\n",
"\u001b[0m\u001b[36;1m\u001b[1;3mSorry, there was an error. Please try again.\u001b[0m\u001b[32;1m\u001b[1;3mParece que hubo un error al intentar calcular el valor de la función mágica. ¿Te gustaría que lo intente de nuevo?\u001b[0m\n",
"\n",
"\u001b[1m> Finished chain.\u001b[0m\n"
]
@@ -752,10 +766,10 @@
"data": {
"text/plain": [
"{'input': 'what is the value of magic_function(3)?',\n",
" 'output': 'Aún no puedo obtener el valor de `magic_function(3)`. ¿Hay algo más en lo que pueda ayudarte?'}"
" 'output': 'Parece que hubo un error al intentar calcular el valor de la función mágica. ¿Te gustaría que lo intente de nuevo?'}"
]
},
"execution_count": 15,
"execution_count": 17,
"metadata": {},
"output_type": "execute_result"
}
@@ -797,7 +811,7 @@
},
{
"cell_type": "code",
"execution_count": 16,
"execution_count": 18,
"id": "b974a91f-6ae8-4644-83d9-73666258a6db",
"metadata": {},
"outputs": [
@@ -805,12 +819,12 @@
"name": "stdout",
"output_type": "stream",
"text": [
"('human', 'what is the value of magic_function(3)?')\n",
"content='Sorry, there was an error. Please try again.' name='magic_function' id='1a08b883-9c7b-4969-9e9b-67ce64cdcb5f' tool_call_id='call_pFdKcCu5taDTtOOfX14vEDRp'\n",
"content='It seems there was an error when trying to apply the magic function. Let me try again.' additional_kwargs={'tool_calls': [{'id': 'call_DA0lpDIkBFg2GHy4WsEcZG4K', 'function': {'arguments': '{\"input\":\"3\"}', 'name': 'magic_function'}, 'type': 'function'}]} response_metadata={'token_usage': {'completion_tokens': 34, 'prompt_tokens': 97, 'total_tokens': 131}, 'model_name': 'gpt-4o', 'system_fingerprint': 'fp_729ea513f7', 'finish_reason': 'tool_calls', 'logprobs': None} id='run-d571b774-0ea3-4e35-8b7d-f32932c3f3cc-0' tool_calls=[{'name': 'magic_function', 'args': {'input': '3'}, 'id': 'call_DA0lpDIkBFg2GHy4WsEcZG4K'}]\n",
"content='Sorry, there was an error. Please try again.' name='magic_function' id='0b45787b-c82a-487f-9a5a-de129c30460f' tool_call_id='call_DA0lpDIkBFg2GHy4WsEcZG4K'\n",
"content='It appears that there is a consistent issue when trying to apply the magic function to the input \"3.\" This could be due to various reasons, such as the input not being in the correct format or an internal error.\\n\\nIf you have any other questions or if there\\'s something else you\\'d like to try, please let me know!' response_metadata={'token_usage': {'completion_tokens': 66, 'prompt_tokens': 153, 'total_tokens': 219}, 'model_name': 'gpt-4o', 'system_fingerprint': 'fp_729ea513f7', 'finish_reason': 'stop', 'logprobs': None} id='run-50a962e6-21b7-4327-8dea-8e2304062627-0'\n"
"content='what is the value of magic_function(3)?' id='74e2d5e8-2b59-4820-979c-8d11ecfc14c2'\n",
"content='Sorry, there was an error. Please try again.' name='magic_function' id='8c37c19b-3586-46b1-aab9-a045786801a2' tool_call_id='call_ihtrH6IG95pDXpKluIwAgi3J'\n",
"content='It seems there was an error in processing the request. Let me try again.' additional_kwargs={'tool_calls': [{'id': 'call_iF0vYWAd6rfely0cXSqdMOnF', 'function': {'arguments': '{\"input\":\"3\"}', 'name': 'magic_function'}, 'type': 'function'}]} response_metadata={'token_usage': {'completion_tokens': 31, 'prompt_tokens': 88, 'total_tokens': 119}, 'model_name': 'gpt-4o-2024-05-13', 'system_fingerprint': 'fp_4e2b2da518', 'finish_reason': 'tool_calls', 'logprobs': None} id='run-eb88ec77-d492-43a5-a5dd-4cefef9a6920-0' tool_calls=[{'name': 'magic_function', 'args': {'input': '3'}, 'id': 'call_iF0vYWAd6rfely0cXSqdMOnF', 'type': 'tool_call'}] usage_metadata={'input_tokens': 88, 'output_tokens': 31, 'total_tokens': 119}\n",
"content='Sorry, there was an error. Please try again.' name='magic_function' id='c9ff261f-a0f1-4c92-a9f2-cd749f62d911' tool_call_id='call_iF0vYWAd6rfely0cXSqdMOnF'\n",
"content='I am currently unable to process the request with the input \"3\" for the `magic_function`. If you have any other questions or need assistance with something else, please let me know!' response_metadata={'token_usage': {'completion_tokens': 39, 'prompt_tokens': 141, 'total_tokens': 180}, 'model_name': 'gpt-4o-2024-05-13', 'system_fingerprint': 'fp_4e2b2da518', 'finish_reason': 'stop', 'logprobs': None} id='run-d42508aa-f286-4b57-80fb-f8a76736d470-0' usage_metadata={'input_tokens': 141, 'output_tokens': 39, 'total_tokens': 180}\n"
]
}
],
@@ -847,7 +861,7 @@
},
{
"cell_type": "code",
"execution_count": 17,
"execution_count": 19,
"id": "4b8498fc-a7af-4164-a401-d8714f082306",
"metadata": {},
"outputs": [
@@ -874,7 +888,7 @@
" 'output': 'Agent stopped due to max iterations.'}"
"content='Sorry there was an error, please try again.' name='magic_function' id='00d5386f-eb23-4628-9a29-d9ce6a7098cc' tool_call_id='call_bTURmOn9C8zslmn0kMFeykIn'\n",
"content='Sorry there was an error, please try again.' name='magic_function' id='ef8ddf1d-9ad7-4ac0-b784-b673c4d94bbd' tool_call_id='call_ujE0IQBbIQnxcF9gsZXQfdhF'\n",
"content='It seems there was an issue with the previous attempt. Let me try that again.' additional_kwargs={'tool_calls': [{'id': 'call_GcsAfCFUHJ50BN2IOWnwTbQ7', 'function': {'arguments': '{\"input\":3}', 'name': 'magic_function'}, 'type': 'function'}]} response_metadata={'token_usage': {'completion_tokens': 32, 'prompt_tokens': 87, 'total_tokens': 119}, 'model_name': 'gpt-4o-2024-05-13', 'system_fingerprint': 'fp_4e2b2da518', 'finish_reason': 'tool_calls', 'logprobs': None} id='run-54527c4b-8ff0-4ee8-8abf-224886bd222e-0' tool_calls=[{'name': 'magic_function', 'args': {'input': 3}, 'id': 'call_GcsAfCFUHJ50BN2IOWnwTbQ7', 'type': 'tool_call'}] usage_metadata={'input_tokens': 87, 'output_tokens': 32, 'total_tokens': 119}\n",
"{'input': 'what is the value of magic_function(3)?', 'output': 'Agent stopped due to max iterations.'}\n"
]
}
@@ -1118,7 +1132,7 @@
},
{
"cell_type": "code",
"execution_count": 22,
"execution_count": 24,
"id": "b94bb169",
"metadata": {},
"outputs": [
@@ -1216,12 +1230,12 @@
"source": [
"### In LangGraph\n",
"\n",
"We can use the [`messages_modifier`](https://langchain-ai.github.io/langgraph/reference/prebuilt/#create_react_agent) just as before when passing in [prompt templates](#prompt-templates)."
"We can use the [`state_modifier`](https://langchain-ai.github.io/langgraph/reference/prebuilt/#create_react_agent) just as before when passing in [prompt templates](#prompt-templates)."
"LCEL is designed to streamline the process of building useful apps with LLMs and combining related components. It does this by providing:\n",
"\n",
"1. **A unified interface**: Every LCEL object implements the `Runnable` interface, which defines a common set of invocation methods (`invoke`, `batch`, `stream`, `ainvoke`, ...). This makes it possible to also automatically and consistently support useful operations like streaming of intermediate steps and batching, since every chain composed of LCEL objects is itself an LCEL object.\n",
"2. **Composition primitives**: LCEL provides a number of primitives that make it easy to compose chains, parallelize components, add fallbacks, dynamically configure chain internals, and more.\n",
"\n",
"LangChain maintains a number of legacy abstractions. Many of these can be reimplemented via short combinations of LCEL primitives. Doing so confers some general advantages:\n",
"\n",
"- The resulting chains typically implement the full `Runnable` interface, including streaming and asynchronous support where appropriate;\n",
"- The chains may be more easily extended or modified;\n",
"- The parameters of the chain are typically surfaced for easier customization (e.g., prompts) over previous versions, which tended to be subclasses and had opaque parameters and internals.\n",
"\n",
"The LCEL implementations can be slightly more verbose, but there are significant benefits in transparency and customizability.\n",
"\n",
"In this guide we review LCEL implementations of common legacy abstractions. Where appropriate, we link out to separate guides with more detail."
"[`LLMChain`](https://api.python.langchain.com/en/latest/chains/langchain.chains.llm.LLMChain.html) combined a prompt template, LLM, and output parser into a class.\n",
"\n",
"Some advantages of switching to the LCEL implementation are:\n",
"\n",
"- Clarity around contents and parameters. The legacy `LLMChain` contains a default output parser and other options.\n",
"- Easier streaming. `LLMChain` only supports streaming via callbacks.\n",
"- Easier access to raw message outputs if desired. `LLMChain` only exposes these via a parameter or via callback.\n",
"\n",
"import { ColumnContainer, Column } from \"@theme/Columns\";\n",
"\n",
"<ColumnContainer>\n",
"\n",
"<Column>\n",
"\n",
"#### Legacy\n"
]
},
{
"cell_type": "code",
"execution_count": 11,
"id": "e628905c-430e-4e4a-9d7c-c91d2f42052e",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'adjective': 'funny',\n",
" 'text': \"Why couldn't the bicycle find its way home?\\n\\nBecause it lost its bearings!\"}"
"Note that `LLMChain` by default returns a `dict` containing both the input and the output. If this behavior is desired, we can replicate it using another LCEL primitive, [`RunnablePassthrough`](https://api.python.langchain.com/en/latest/runnables/langchain_core.runnables.passthrough.RunnablePassthrough.html):"
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "529206c5-abbe-4213-9e6c-3b8586c8000d",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'adjective': 'funny',\n",
" 'text': \"Why couldn't the bicycle stand up by itself?\\n\\nBecause it was two tired!\"}"
"[`ConversationChain`](https://api.python.langchain.com/en/latest/chains/langchain.chains.conversation.base.ConversationChain.html) incorporates a memory of previous messages to sustain a stateful conversation.\n",
"\n",
"Some advantages of switching to the LCEL implementation are:\n",
"\n",
"- Innate support for threads/separate sessions. To make this work with `ConversationChain`, you'd need to instantiate a separate memory class outside the chain.\n",
"- More explicit parameters. `ConversationChain` contains a hidden default prompt, which can cause confusion.\n",
"- Streaming support. `ConversationChain` only supports streaming via callbacks.\n",
"\n",
"`RunnableWithMessageHistory` implements sessions via configuration parameters. It should be instantiated with a callable that returns a [chat message history](https://api.python.langchain.com/en/latest/chat_history/langchain_core.chat_history.BaseChatMessageHistory.html). By default, it expects this function to take a single argument `session_id`.\n",
"\n",
"<ColumnContainer>\n",
"<Column>\n",
"\n",
"#### Legacy\n"
]
},
{
"cell_type": "code",
"execution_count": 15,
"id": "4f2cc6dc-d70a-4c13-9258-452f14290da6",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'input': 'how are you?',\n",
" 'history': '',\n",
" 'response': \"Arrr, I be doin' well, me matey! Just sailin' the high seas in search of treasure and adventure. How can I assist ye today?\"}"
"See [this tutorial](/docs/tutorials/chatbot) for a more end-to-end guide on building with [`RunnableWithMessageHistory`](https://api.python.langchain.com/en/latest/runnables/langchain_core.runnables.history.RunnableWithMessageHistory.html).\n",
"The [`RetrievalQA`](https://api.python.langchain.com/en/latest/chains/langchain.chains.retrieval_qa.base.RetrievalQA.html) chain performed natural-language question answering over a data source using retrieval-augmented generation.\n",
"\n",
"Some advantages of switching to the LCEL implementation are:\n",
"\n",
"- Easier customizability. Details such as the prompt and how documents are formatted are only configurable via specific parameters in the `RetrievalQA` chain.\n",
"- More easily return source documents.\n",
"- Support for runnable methods like streaming and async operations.\n",
"\n",
"Now let's look at them side-by-side. We'll use the same ingestion code to load a [blog post by Lilian Weng](https://lilianweng.github.io/posts/2023-06-23-agent/) on autonomous agents into a local vector store:"
" 'result': 'Autonomous agents are LLM-empowered agents that handle autonomous design, planning, and performance of complex tasks, such as scientific experiments. These agents can browse the Internet, read documentation, execute code, call robotics experimentation APIs, and leverage other LLMs. They are capable of reasoning and planning ahead for complicated tasks by breaking them down into smaller steps.'}"
]
},
"execution_count": 22,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"from langchain import hub\n",
"from langchain.chains import RetrievalQA\n",
"\n",
"# See full prompt at https://smith.langchain.com/hub/rlm/rag-prompt\n",
"'Autonomous agents are agents that can handle autonomous design, planning, and performance of complex tasks, such as scientific experiments. They can browse the Internet, read documentation, execute code, call robotics experimentation APIs, and leverage other language model models. These agents use reasoning steps to develop solutions to specific tasks, like creating a novel anticancer drug.'"
"qa_chain.invoke(\"What are autonomous agents?\")"
]
},
{
"cell_type": "markdown",
"id": "d6f44fe8",
"metadata": {},
"source": [
"</Column>\n",
"</ColumnContainer>\n",
"\n",
"The LCEL implementation exposes the internals of what's happening around retrieving, formatting documents, and passing them through a prompt to the LLM, but it is more verbose. You can customize and wrap this composition logic in a helper function, or use the higher-level [`create_retrieval_chain`](https://api.python.langchain.com/en/latest/chains/langchain.chains.retrieval.create_retrieval_chain.html) and [`create_stuff_documents_chain`](https://api.python.langchain.com/en/latest/chains/langchain.chains.combine_documents.stuff.create_stuff_documents_chain.html) helper method:"
]
},
{
"cell_type": "code",
"execution_count": 20,
"id": "5fe42761",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'input': 'What are autonomous agents?',\n",
" 'context': [Document(page_content='Boiko et al. (2023) also looked into LLM-empowered agents for scientific discovery, to handle autonomous design, planning, and performance of complex scientific experiments. This agent can use tools to browse the Internet, read documentation, execute code, call robotics experimentation APIs and leverage other LLMs.\\nFor example, when requested to \"develop a novel anticancer drug\", the model came up with the following reasoning steps:', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'}),\n",
" Document(page_content='Weng, Lilian. (Jun 2023). “LLM-powered Autonomous Agents”. Lil’Log. https://lilianweng.github.io/posts/2023-06-23-agent/.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'}),\n",
" Document(page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\\nComponent One: Planning#\\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\\nTask Decomposition#', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'}),\n",
" Document(page_content=\"LLM Powered Autonomous Agents | Lil'Log\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\nLil'Log\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n\\nPosts\\n\\n\\n\\n\\nArchive\\n\\n\\n\\n\\nSearch\\n\\n\\n\\n\\nTags\\n\\n\\n\\n\\nFAQ\\n\\n\\n\\n\\nemojisearch.app\\n\\n\\n\\n\\n\\n\\n\\n\\n\\n LLM Powered Autonomous Agents\\n \\nDate: June 23, 2023 | Estimated Reading Time: 31 min | Author: Lilian Weng\\n\\n\\n \\n\\n\\nTable of Contents\\n\\n\\n\\nAgent System Overview\\n\\nComponent One: Planning\\n\\nTask Decomposition\\n\\nSelf-Reflection\\n\\n\\nComponent Two: Memory\\n\\nTypes of Memory\\n\\nMaximum Inner Product Search (MIPS)\", metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'})],\n",
" 'answer': 'Autonomous agents are entities that can operate independently, making decisions and taking actions without direct human intervention. These agents can perform tasks such as planning, executing complex experiments, and leveraging various tools and resources to achieve objectives. In the context provided, LLM-powered autonomous agents are specifically designed for scientific discovery, capable of handling tasks like designing novel anticancer drugs through reasoning steps.'}"
"The [`ConversationalRetrievalChain`](https://api.python.langchain.com/en/latest/chains/langchain.chains.conversational_retrieval.base.ConversationalRetrievalChain.html) was an all-in one way that combined retrieval-augmented generation with chat history, allowing you to \"chat with\" your documents.\n",
"\n",
"Advantages of switching to the LCEL implementation are similar to the `RetrievalQA` section above:\n",
"\n",
"- Clearer internals. The `ConversationalRetrievalChain` chain hides an entire question rephrasing step which dereferences the initial query against the chat history.\n",
" - This means the class contains two sets of configurable prompts, LLMs, etc.\n",
"- More easily return source documents.\n",
"- Support for runnable methods like streaming and async operations.\n",
"\n",
"Here are side-by-side implementations with custom prompts. We'll reuse the loaded documents and vector store from the previous section:"
]
},
{
"cell_type": "markdown",
"id": "8bc06416",
"metadata": {},
"source": [
"<ColumnContainer>\n",
"\n",
"<Column>\n",
"\n",
"#### Legacy"
]
},
{
"cell_type": "code",
"execution_count": 31,
"id": "54eb9576",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'question': 'What are autonomous agents?',\n",
" 'chat_history': '',\n",
" 'answer': 'Autonomous agents are powered by Large Language Models (LLMs) to handle tasks like scientific discovery and complex experiments autonomously. These agents can browse the internet, read documentation, execute code, and leverage other LLMs to perform tasks. They can reason and plan ahead to decompose complicated tasks into manageable steps.'}"
" \"question\": \"What are autonomous agents?\",\n",
" \"chat_history\": \"\",\n",
" }\n",
")"
]
},
{
"cell_type": "markdown",
"id": "43a8a23c",
"metadata": {},
"source": [
"</Column>\n",
"\n",
"<Column>\n",
"\n",
"#### LCEL\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": 25,
"id": "c884b138",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'input': 'What are autonomous agents?',\n",
" 'chat_history': [],\n",
" 'context': [Document(page_content='Boiko et al. (2023) also looked into LLM-empowered agents for scientific discovery, to handle autonomous design, planning, and performance of complex scientific experiments. This agent can use tools to browse the Internet, read documentation, execute code, call robotics experimentation APIs and leverage other LLMs.\\nFor example, when requested to \"develop a novel anticancer drug\", the model came up with the following reasoning steps:', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'}),\n",
" Document(page_content='Weng, Lilian. (Jun 2023). “LLM-powered Autonomous Agents”. Lil’Log. https://lilianweng.github.io/posts/2023-06-23-agent/.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'}),\n",
" Document(page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\\nComponent One: Planning#\\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\\nTask Decomposition#', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'}),\n",
" Document(page_content='Or\\n@article{weng2023agent,\\n title = \"LLM-powered Autonomous Agents\",\\n author = \"Weng, Lilian\",\\n journal = \"lilianweng.github.io\",\\n year = \"2023\",\\n month = \"Jun\",\\n url = \"https://lilianweng.github.io/posts/2023-06-23-agent/\"\\n}\\nReferences#\\n[1] Wei et al. “Chain of thought prompting elicits reasoning in large language models.” NeurIPS 2022\\n[2] Yao et al. “Tree of Thoughts: Dliberate Problem Solving with Large Language Models.” arXiv preprint arXiv:2305.10601 (2023).', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/', 'title': \"LLM Powered Autonomous Agents | Lil'Log\", 'description': 'Building agents with LLM (large language model) as its core controller is a cool concept. Several proof-of-concepts demos, such as AutoGPT, GPT-Engineer and BabyAGI, serve as inspiring examples. The potentiality of LLM extends beyond generating well-written copies, stories, essays and programs; it can be framed as a powerful general problem solver.\\nAgent System Overview In a LLM-powered autonomous agent system, LLM functions as the agent’s brain, complemented by several key components:', 'language': 'en'})],\n",
" 'answer': 'Autonomous agents are entities capable of acting independently, making decisions, and performing tasks without direct human intervention. These agents can interact with their environment, perceive information, and take actions based on their goals or objectives. They often use artificial intelligence techniques to navigate and accomplish tasks in complex or dynamic environments.'}"
# How to use LangChain with different Pydantic versions
- Pydantic v2 was released in June, 2023 (https://docs.pydantic.dev/2.0/blog/pydantic-v2-final/)
- v2 contains has a number of breaking changes (https://docs.pydantic.dev/2.0/migration/)
- Pydantic v2 and v1 are under the same package name, so both versions cannot be installed at the same time
- Pydantic v2 was released in June, 2023 (https://docs.pydantic.dev/2.0/blog/pydantic-v2-final/).
- v2 contains has a number of breaking changes (https://docs.pydantic.dev/2.0/migration/).
- Pydantic 1 End of Life was in June 2024. LangChain will be dropping support for Pydantic 1 in the near future,
and likely migrating internally to Pydantic 2. The timeline is tentatively September. This change will be accompanied by a minor version bump in the main langchain packages to version 0.3.x.
## LangChain Pydantic migration plan
As of `langchain>=0.0.267`, LangChain allows users to install either Pydantic V1 or V2.
As of `langchain>=0.0.267`, LangChain will allow users to install either Pydantic V1 or V2.
* Internally LangChain will continue to [use V1](https://docs.pydantic.dev/latest/migration/#continue-using-pydantic-v1-features).
* During this time, users can pin their pydantic version to v1 to avoid breaking changes, or start a partial
migration using pydantic v2 throughout their code, but avoiding mixing v1 and v2 code for LangChain (see below).
Internally, LangChain continues to use the [Pydantic V1](https://docs.pydantic.dev/latest/migration/#continue-using-pydantic-v1-features) via
the v1 namespace of Pydantic 2.
User can either pin to pydantic v1, and upgrade their code in one go once LangChain has migrated to v2 internally, or they can start a partial migration to v2, but must avoid mixing v1 and v2 code for LangChain.
Because Pydantic does not support mixing .v1 and .v2 objects, users should be aware of a number of issues
when using LangChain with Pydantic.
:::caution
While LangChain supports Pydantic V2 objects in some APIs (listed below), it's suggested that users keep using Pydantic V1 objects until LangChain 0.3 is released.
:::
## 1. Passing Pydantic objects to LangChain APIs
Most LangChain APIs for *tool usage* (see list below) have been updated to accept either Pydantic v1 or v2 objects.
* Pydantic v1 objects correspond to subclasses of `pydantic.BaseModel` if `pydantic 1` is installed or subclasses of `pydantic.v1.BaseModel` if `pydantic 2` is installed.
* Pydantic v2 objects correspond to subclasses of `pydantic.BaseModel` if `pydantic 2` is installed.
"1. Using the built-in [create_retrieval_chain](https://api.python.langchain.com/en/latest/chains/langchain.chains.retrieval.create_retrieval_chain.html), which returns sources by default;\n",
"2. Using a simple [LCEL](/docs/concepts#langchain-expression-language-lcel) implementation, to show the operating principle."
"2. Using a simple [LCEL](/docs/concepts#langchain-expression-language-lcel) implementation, to show the operating principle.\n",
"\n",
"We will also show how to structure sources into the model response, such that a model can report what specific sources it used in generating its answer."
]
},
{
@@ -130,8 +132,8 @@
},
{
"cell_type": "code",
"execution_count": 3,
"id": "820244ae-74b4-4593-b392-822979dd91b8",
"execution_count": null,
"id": "24a69b8c-024e-4e34-b827-9c9de46512a3",
"metadata": {},
"outputs": [],
"source": [
@@ -211,11 +213,11 @@
"data": {
"text/plain": [
"{'input': 'What is Task Decomposition?',\n",
" 'context': [Document(page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\\nComponent One: Planning#\\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\\nTask Decomposition#\\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking process.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}),\n",
" Document(page_content='Tree of Thoughts (Yao et al. 2023) extends CoT by exploring multiple reasoning possibilities at each step. It first decomposes the problem into multiple thought steps and generates multiple thoughts per step, creating a tree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\\nTask decomposition can be done (1) by LLM with simple prompting like \"Steps for XYZ.\\\\n1.\", \"What are the subgoals for achieving XYZ?\", (2) by using task-specific instructions; e.g. \"Write a story outline.\" for writing a novel, or (3) with human inputs.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}),\n",
" Document(page_content='Resources:\\n1. Internet access for searches and information gathering.\\n2. Long Term memory management.\\n3. GPT-3.5 powered Agents for delegation of simple tasks.\\n4. File output.\\n\\nPerformance Evaluation:\\n1. Continuously review and analyze your actions to ensure you are performing to the best of your abilities.\\n2. Constructively self-criticize your big-picture behavior constantly.\\n3. Reflect on past decisions and strategies to refine your approach.\\n4. Every command has a cost, so be smart and efficient. Aim to complete tasks in the least number of steps.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}),\n",
" Document(page_content=\"(3) Task execution: Expert models execute on the specific tasks and log results.\\nInstruction:\\n\\nWith the input and the inference results, the AI assistant needs to describe the process and results. The previous stages can be formed as - User Input: {{ User Input }}, Task Planning: {{ Tasks }}, Model Selection: {{ Model Assignment }}, Task Execution: {{ Predictions }}. You must first answer the user's request in a straightforward manner. Then describe the task process and show your analysis and model inference results to the user in the first person. If inference results contain a file path, must tell the user the complete file path.\", metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'})],\n",
" 'answer': 'Task decomposition involves breaking down a complex task into smaller and simpler steps. This process helps agents or models handle challenging tasks by dividing them into more manageable subtasks. Techniques like Chain of Thought and Tree of Thoughts are used to decompose tasks into multiple steps for better problem-solving.'}"
" 'context': [Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\\nComponent One: Planning#\\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\\nTask Decomposition#\\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking process.'),\n",
" Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content='Tree of Thoughts (Yao et al. 2023) extends CoT by exploring multiple reasoning possibilities at each step. It first decomposes the problem into multiple thought steps and generates multiple thoughts per step, creating a tree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\\nTask decomposition can be done (1) by LLM with simple prompting like \"Steps for XYZ.\\\\n1.\", \"What are the subgoals for achieving XYZ?\", (2) by using task-specific instructions; e.g. \"Write a story outline.\" for writing a novel, or (3) with human inputs.'),\n",
" Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content='Resources:\\n1. Internet access for searches and information gathering.\\n2. Long Term memory management.\\n3. GPT-3.5 powered Agents for delegation of simple tasks.\\n4. File output.\\n\\nPerformance Evaluation:\\n1. Continuously review and analyze your actions to ensure you are performing to the best of your abilities.\\n2. Constructively self-criticize your big-picture behavior constantly.\\n3. Reflect on past decisions and strategies to refine your approach.\\n4. Every command has a cost, so be smart and efficient. Aim to complete tasks in the least number of steps.'),\n",
" Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content=\"(3) Task execution: Expert models execute on the specific tasks and log results.\\nInstruction:\\n\\nWith the input and the inference results, the AI assistant needs to describe the process and results. The previous stages can be formed as - User Input: {{ User Input }}, Task Planning: {{ Tasks }}, Model Selection: {{ Model Assignment }}, Task Execution: {{ Predictions }}. You must first answer the user's request in a straightforward manner. Then describe the task process and show your analysis and model inference results to the user in the first person. If inference results contain a file path, must tell the user the complete file path.\")],\n",
" 'answer': 'Task decomposition involves breaking down a complex task into smaller and more manageable steps. This process helps agents or models tackle difficult tasks by dividing them into simpler subtasks or components. Task decomposition can be achieved through techniques like Chain of Thought or Tree of Thoughts, which guide the agent in breaking down tasks into sequential or branching steps.'}"
]
},
"execution_count": 5,
@@ -251,18 +253,18 @@
{
"cell_type": "code",
"execution_count": 6,
"id": "22ea137c-1a7a-44dd-ac73-281213979957",
"id": "1950953a-e6f1-439d-b7b9-c3bd456e388d",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'input': 'What is Task Decomposition',\n",
" 'context': [Document(page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\\nComponent One: Planning#\\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\\nTask Decomposition#\\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking process.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}),\n",
" Document(page_content='Tree of Thoughts (Yao et al. 2023) extends CoT by exploring multiple reasoning possibilities at each step. It first decomposes the problem into multiple thought steps and generates multiple thoughts per step, creating a tree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\\nTask decomposition can be done (1) by LLM with simple prompting like \"Steps for XYZ.\\\\n1.\", \"What are the subgoals for achieving XYZ?\", (2) by using task-specific instructions; e.g. \"Write a story outline.\" for writing a novel, or (3) with human inputs.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}),\n",
" Document(page_content='The AI assistant can parse user input to several tasks: [{\"task\": task, \"id\", task_id, \"dep\": dependency_task_ids, \"args\": {\"text\": text, \"image\": URL, \"audio\": URL, \"video\": URL}}]. The \"dep\" field denotes the id of the previous task which generates a new resource that the current task relies on. A special tag \"-task_id\" refers to the generated text image, audio and video in the dependency task with id as task_id. The task MUST be selected from the following options: {{ Available Task List }}. There is a logical relationship between tasks, please note their order. If the user input can\\'t be parsed, you need to reply empty JSON. Here are several cases for your reference: {{ Demonstrations }}. The chat history is recorded as {{ Chat History }}. From this chat history, you can find the path of the user-mentioned resources for your task planning.', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}),\n",
" Document(page_content='Fig. 11. Illustration of how HuggingGPT works. (Image source: Shen et al. 2023)\\nThe system comprises of 4 stages:\\n(1) Task planning: LLM works as the brain and parses the user requests into multiple tasks. There are four attributes associated with each task: task type, ID, dependencies, and arguments. They use few-shot examples to guide LLM to do task parsing and planning.\\nInstruction:', metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'})],\n",
" 'answer': 'Task decomposition involves breaking down complex tasks into smaller and simpler steps to make them more manageable for autonomous agents or models. This process can be achieved by techniques like Chain of Thought (CoT) or Tree of Thoughts, which guide the model to think step by step or explore multiple reasoning possibilities at each step. Task decomposition can be done through simple prompting with language models, task-specific instructions, or human inputs.'}"
" 'context': [Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content='Fig. 1. Overview of a LLM-powered autonomous agent system.\\nComponent One: Planning#\\nA complicated task usually involves many steps. An agent needs to know what they are and plan ahead.\\nTask Decomposition#\\nChain of thought (CoT; Wei et al. 2022) has become a standard prompting technique for enhancing model performance on complex tasks. The model is instructed to “think step by step” to utilize more test-time computation to decompose hard tasks into smaller and simpler steps. CoT transforms big tasks into multiple manageable tasks and shed lights into an interpretation of the model’s thinking process.'),\n",
" Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content='Tree of Thoughts (Yao et al. 2023) extends CoT by exploring multiple reasoning possibilities at each step. It first decomposes the problem into multiple thought steps and generates multiple thoughts per step, creating a tree structure. The search process can be BFS (breadth-first search) or DFS (depth-first search) with each state evaluated by a classifier (via a prompt) or majority vote.\\nTask decomposition can be done (1) by LLM with simple prompting like \"Steps for XYZ.\\\\n1.\", \"What are the subgoals for achieving XYZ?\", (2) by using task-specific instructions; e.g. \"Write a story outline.\" for writing a novel, or (3) with human inputs.'),\n",
" Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content='The AI assistant can parse user input to several tasks: [{\"task\": task, \"id\", task_id, \"dep\": dependency_task_ids, \"args\": {\"text\": text, \"image\": URL, \"audio\": URL, \"video\": URL}}]. The \"dep\" field denotes the id of the previous task which generates a new resource that the current task relies on. A special tag \"-task_id\" refers to the generated text image, audio and video in the dependency task with id as task_id. The task MUST be selected from the following options: {{ Available Task List }}. There is a logical relationship between tasks, please note their order. If the user input can\\'t be parsed, you need to reply empty JSON. Here are several cases for your reference: {{ Demonstrations }}. The chat history is recorded as {{ Chat History }}. From this chat history, you can find the path of the user-mentioned resources for your task planning.'),\n",
" Document(metadata={'source': 'https://lilianweng.github.io/posts/2023-06-23-agent/'}, page_content='Fig. 11. Illustration of how HuggingGPT works. (Image source: Shen et al. 2023)\\nThe system comprises of 4 stages:\\n(1) Task planning: LLM works as the brain and parses the user requests into multiple tasks. There are four attributes associated with each task: task type, ID, dependencies, and arguments. They use few-shot examples to guide LLM to do task parsing and planning.\\nInstruction:')],\n",
" 'answer': 'Task decomposition is a technique used in artificial intelligence to break down complex tasks into smaller and more manageable subtasks. This approach helps agents or models to tackle difficult problems by dividing them into simpler steps, improving performance and interpretability. Different methods like Chain of Thought and Tree of Thoughts have been developed to enhance task decomposition in AI systems.'}"
]
},
"execution_count": 6,
@@ -279,15 +281,25 @@
" return \"\\n\\n\".join(doc.page_content for doc in docs)\n",
"\n",
"\n",
"# This Runnable takes a dict with keys 'input' and 'context',\n",
"# formats them into a prompt, and generates a response.\n",
"Check out the [LangSmith trace](https://smith.langchain.com/public/0cb42685-e29e-4280-a503-bef2014d7ba2/r)\n",
"Check out the [LangSmith trace](https://smith.langchain.com/public/1c055a3b-0236-4670-a3fb-023d418ba796/r)\n",
"\n",
":::"
]
},
{
"cell_type": "markdown",
"id": "c1c17797-d965-4fd2-b8d4-d386f25dd352",
"metadata": {},
"source": [
"## Structure sources in model response\n",
"\n",
"Up to this point, we've simply propagated the documents returned from the retrieval step through to the final response. But this may not illustrate what subset of information the model relied on when generating its answer. Below, we show how to structure sources into the model response, allowing the model to report what specific context it relied on for its answer.\n",
"\n",
"Because the above LCEL implementation is composed of [Runnable](/docs/concepts/#runnable-interface) primitives, it is straightforward to extend. Below, we make a simple change:\n",
"\n",
"- We use the model's tool-calling features to generate [structured output](/docs/how_to/structured_output/), consisting of an answer and list of sources. The schema for the response is represented in the `AnswerWithSources` TypedDict, below.\n",
"- We remove the `StrOutputParser()`, as we expect `dict` output in this scenario."
"response = chain.invoke({\"input\": \"What is Chain of Thought?\"})"
]
},
{
"cell_type": "code",
"execution_count": 18,
"id": "7a8fc0c5-afb3-4012-a467-3951996a6850",
"metadata": {},
"outputs": [
{
"name": "stdout",
"output_type": "stream",
"text": [
"{\n",
" \"answer\": \"Chain of Thought (CoT) is a prompting technique that enhances model performance on complex tasks by instructing the model to \\\"think step by step\\\" to decompose hard tasks into smaller and simpler steps. It transforms big tasks into multiple manageable tasks and sheds light on the interpretation of the model's thinking process.\",\n",
"We can pass in secrets to our runnables at runtime using the `RunnableConfig`. Specifically we can pass in secrets with a `__` prefix to the `configurable` field. This will ensure that these secrets aren't traced as part of the invocation:"
"Looking at the LangSmith trace for this run, we can see that \"traced_key\" was recorded (as part of Metadata) while our secret int was not: https://smith.langchain.com/public/aa7e3289-49ca-422d-a408-f6b927210170/r"
"In this method, the gradient of distance is used to split chunks along with the percentile method.\n",
"This method is useful when chunks are highly correlated with each other or specific to a domain e.g. legal or medical. The idea is to apply anomaly detection on gradient array so that the distribution become wider and easy to identify boundaries in highly semantic data."
"* [SQL tutorial](/docs/tutorials/sql_qa): Many of the challenges of working with SQL db's and CSV's are generic to any structured data type, so it's useful to read the SQL techniques even if you're using Pandas for CSV data analysis.\n",
"* [Tool use](/docs/how_to/tool_calling): Guides on general best practices when working with chains and agents that invoke tools\n",
"* [Agents](/docs/tutorials/agents): Understand the fundamentals of building LLM agents.\n",
"* Integrations: Sandboxed envs like [E2B](/docs/integrations/tools/e2b_data_analysis) and [Bearly](/docs/integrations/tools/bearly), utilities like [SQLDatabase](https://api.python.langchain.com/en/latest/utilities/langchain_community.utilities.sql_database.SQLDatabase.html#langchain_community.utilities.sql_database.SQLDatabase), related agents like [Spark DataFrame agent](/docs/integrations/toolkits/spark)."
"* Integrations: Sandboxed envs like [E2B](/docs/integrations/tools/e2b_data_analysis) and [Bearly](/docs/integrations/tools/bearly), utilities like [SQLDatabase](https://api.python.langchain.com/en/latest/utilities/langchain_community.utilities.sql_database.SQLDatabase.html#langchain_community.utilities.sql_database.SQLDatabase), related agents like [Spark DataFrame agent](/docs/integrations/tools/spark_sql)."
"This is the easiest and most reliable way to get structured outputs. `with_structured_output()` is implemented for models that provide native APIs for structuring outputs, like tool/function calling or JSON mode, and makes use of these capabilities under the hood.\n",
"\n",
"This method takes a schema as input which specifies the names, types, and descriptions of the desired output attributes. The method returns a model-like Runnable, except that instead of outputting strings or Messages it outputs objects corresponding to the given schema. The schema can be specified as a [JSON Schema](https://json-schema.org/) or a Pydantic class. If JSON Schema is used then a dictionary will be returned by the Runnable, and if a Pydantic class is used then Pydantic objects will be returned.\n",
"This method takes a schema as input which specifies the names, types, and descriptions of the desired output attributes. The method returns a model-like Runnable, except that instead of outputting strings or Messages it outputs objects corresponding to the given schema. The schema can be specified as a TypedDict class, [JSON Schema](https://json-schema.org/) or a Pydantic class. If TypedDict or JSON Schema are used then a dictionary will be returned by the Runnable, and if a Pydantic class is used then a Pydantic object will be returned.\n",
"\n",
"As an example, let's get a model to generate a joke and separate the setup from the punchline:\n",
"If we want the model to return a Pydantic object, we just need to pass in the desired Pydantic class:"
"### Pydantic class\n",
"\n",
"If we want the model to return a Pydantic object, we just need to pass in the desired Pydantic class. The key advantage of using Pydantic is that the model-generated output will be validated. Pydantic will raise an error if any required fields are missing or if any fields are of the wrong type."
]
},
{
"cell_type": "code",
"execution_count": 3,
"execution_count": 4,
"id": "070bf702",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"Joke(setup='Why was the cat sitting on the computer?', punchline='Because it wanted to keep an eye on the mouse!', rating=8)"
"Joke(setup='Why was the cat sitting on the computer?', punchline='Because it wanted to keep an eye on the mouse!', rating=7)"
"We can also pass in a [JSON Schema](https://json-schema.org/) dict if you prefer not to use Pydantic. In this case, the response is also a dict:"
"### TypedDict or JSON Schema\n",
"\n",
"If you don't want to use Pydantic, explicitly don't want validation of the arguments, or want to be able to stream the model outputs, you can define your schema using a TypedDict class. We can optionally use a special `Annotated` syntax supported by LangChain that allows you to specify the default value and description of a field. Note, the default value is *not* filled in automatically if the model doesn't generate it, it is only used in defining the schema that is passed to the model.\n",
"\n",
":::info Requirements\n",
"\n",
"- Core: `langchain-core>=0.2.26`\n",
"- Typing extensions: It is highly recommended to import `Annotated` and `TypedDict` from `typing_extensions` instead of `typing` to ensure consistent behavior across Python versions.\n",
"\n",
":::"
]
},
{
"cell_type": "code",
"execution_count": 8,
"id": "70d82891-42e8-424a-919e-07d83bcfec61",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'setup': 'Why was the cat sitting on the computer?',\n",
" 'punchline': 'Because it wanted to keep an eye on the mouse!',\n",
"structured_llm.invoke(\"Tell me a joke about cats\")"
]
},
{
"cell_type": "markdown",
"id": "e4d7b4dc-f617-4ea8-aa58-847c228791b4",
"metadata": {},
"source": [
"Equivalently, we can pass in a [JSON Schema](https://json-schema.org/) dict. This requires no imports or classes and makes it very clear exactly how each parameter is documented, at the cost of being a bit more verbose."
]
},
{
"cell_type": "code",
"execution_count": 6,
"id": "6700994a",
"metadata": {},
"outputs": [
@@ -144,10 +210,10 @@
"text/plain": [
"{'setup': 'Why was the cat sitting on the computer?',\n",
" 'punchline': 'Because it wanted to keep an eye on the mouse!',\n",
" 'rating': 8}"
" 'rating': 7}"
]
},
"execution_count": 8,
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
@@ -169,6 +235,7 @@
" \"rating\": {\n",
" \"type\": \"integer\",\n",
" \"description\": \"How funny the joke is, from 1 to 10\",\n",
" \"default\": None,\n",
" },\n",
" },\n",
" \"required\": [\"setup\", \"punchline\"],\n",
@@ -185,7 +252,7 @@
"source": [
"### Choosing between multiple schemas\n",
"\n",
"The simplest way to let the model choose from multiple schemas is to create a parent Pydantic class that has a Union-typed attribute:"
"The simplest way to let the model choose from multiple schemas is to create a parent schema that has a Union-typed attribute:"
]
},
{
@@ -209,6 +276,17 @@
"from typing import Union\n",
"\n",
"\n",
"# Pydantic\n",
"class Joke(BaseModel):\n",
" \"\"\"Joke to tell user.\"\"\"\n",
"\n",
" setup: str = Field(description=\"The setup of the joke\")\n",
" punchline: str = Field(description=\"The punchline to the joke\")\n",
" rating: Optional[int] = Field(\n",
" default=None, description=\"How funny the joke is, from 1 to 10\"\n",
" )\n",
"\n",
"\n",
"class ConversationalResponse(BaseModel):\n",
" \"\"\"Respond in a conversational manner. Be kind and helpful.\"\"\"\n",
"\n",
@@ -260,7 +338,7 @@
"source": [
"### Streaming\n",
"\n",
"We can stream outputs from our structured model when the output type is a dict (i.e., when the schema is specified as a JSON Schema dict). \n",
"We can stream outputs from our structured model when the output type is a dict (i.e., when the schema is specified as a TypedDict class or JSON Schema dict). \n",
"\n",
":::info\n",
"\n",
@@ -271,7 +349,7 @@
},
{
"cell_type": "code",
"execution_count": 43,
"execution_count": 9,
"id": "aff89877-28a3-472f-a1aa-eff893fe7736",
"metadata": {},
"outputs": [
@@ -302,12 +380,24 @@
"{'setup': 'Why was the cat sitting on the computer?', 'punchline': 'Because it wanted to keep an eye on the'}\n",
"{'setup': 'Why was the cat sitting on the computer?', 'punchline': 'Because it wanted to keep an eye on the mouse'}\n",
"{'setup': 'Why was the cat sitting on the computer?', 'punchline': 'Because it wanted to keep an eye on the mouse!'}\n",
"{'setup': 'Why was the cat sitting on the computer?', 'punchline': 'Because it wanted to keep an eye on the mouse!', 'rating': 8}\n"
"{'setup': 'Why was the cat sitting on the computer?', 'punchline': 'Because it wanted to keep an eye on the mouse!', 'rating': 7}\n"
" \"Tell me a joke about cats, respond in JSON with `setup` and `punchline` keys\"\n",
@@ -526,19 +617,21 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 17,
"id": "10ed2842",
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"{'raw': AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_ASK4EmZeZ69Fi3p554Mb4rWy', 'function': {'arguments': '{\"setup\":\"Why was the cat sitting on the computer?\",\"punchline\":\"Because it wanted to keep an eye on the mouse!\"}', 'name': 'Joke'}, 'type': 'function'}]}, response_metadata={'token_usage': {'completion_tokens': 36, 'prompt_tokens': 107, 'total_tokens': 143}, 'model_name': 'gpt-4-0125-preview', 'system_fingerprint': None, 'finish_reason': 'stop', 'logprobs': None}, id='run-6491d35b-9164-4656-b75c-d7882cfb76cb-0', tool_calls=[{'name': 'Joke', 'args': {'setup': 'Why was the cat sitting on the computer?', 'punchline': 'Because it wanted to keep an eye on the mouse!'}, 'id': 'call_ASK4EmZeZ69Fi3p554Mb4rWy'}], usage_metadata={'input_tokens': 107, 'output_tokens': 36, 'total_tokens': 143}),\n",
" 'parsed': Joke(setup='Why was the cat sitting on the computer?', punchline='Because it wanted to keep an eye on the mouse!', rating=None),\n",
"{'raw': AIMessage(content='', additional_kwargs={'tool_calls': [{'id': 'call_f25ZRmh8u5vHlOWfTUw8sJFZ', 'function': {'arguments': '{\"setup\":\"Why was the cat sitting on the computer?\",\"punchline\":\"Because it wanted to keep an eye on the mouse!\",\"rating\":7}', 'name': 'Joke'}, 'type': 'function'}]}, response_metadata={'token_usage': {'completion_tokens': 33, 'prompt_tokens': 93, 'total_tokens': 126}, 'model_name': 'gpt-4o-2024-05-13', 'system_fingerprint': 'fp_4e2b2da518', 'finish_reason': 'stop', 'logprobs': None}, id='run-d880d7e2-df08-4e9e-ad92-dfc29f2fd52f-0', tool_calls=[{'name': 'Joke', 'args': {'setup': 'Why was the cat sitting on the computer?', 'punchline': 'Because it wanted to keep an eye on the mouse!', 'rating': 7}, 'id': 'call_f25ZRmh8u5vHlOWfTUw8sJFZ', 'type': 'tool_call'}], usage_metadata={'input_tokens': 93, 'output_tokens': 33, 'total_tokens': 126}),\n",
" 'parsed': {'setup': 'Why was the cat sitting on the computer?',\n",
" 'punchline': 'Because it wanted to keep an eye on the mouse!',\n",
"[Tool calling](/docs/concepts/#functiontool-calling) allows a chat model to respond to a given prompt by \"calling a tool\".\n",
"\n",
"Remember, while the name \"tool calling\" implies that the model is directly performing some action, this is actually not the case! The model only generates the arguments to a tool, and actually running the tool (or not) is up to the user.\n",
"\n",
"Tool calling is a general technique that generates structured output from a model, and you can use it even when you don't intend to invoke any tools. An example use-case of that is [extraction from unstructured text](/docs/tutorials/extraction/).\n",
"\n",
"\n",
"\n",
"If you want to see how to use the model-generated tool call to actually run a tool [check out this guide](/docs/how_to/tool_results_pass_to_model/).\n",
"\n",
":::note Supported models\n",
"\n",
"Tool calling is not universal, but is supported by many popular LLM providers. You can find a [list of all models that support tool calling here](/docs/integrations/chat/).\n",
"\n",
":::\n",
"\n",
":::info Tool calling vs function calling\n",
"\n",
"We use the term tool calling interchangeably with function calling. Although\n",
"function calling is sometimes meant to refer to invocations of a single function,\n",
"we treat all models as though they can return multiple tool or function calls in \n",
"each message.\n",
"\n",
":::\n",
"\n",
":::info Supported models\n",
"\n",
"You can find a [list of all models that support tool calling](/docs/integrations/chat/).\n",
"\n",
":::\n",
"\n",
"Tool calling allows a chat model to respond to a given prompt by \"calling a tool\".\n",
"While the name implies that the model is performing \n",
"some action, this is actually not the case! The model generates the \n",
"arguments to a tool, and actually running the tool (or not) is up to the user.\n",
"For example, if you want to [extract output matching some schema](/docs/how_to/structured_output/) \n",
"from unstructured text, you could give the model an \"extraction\" tool that takes \n",
"parameters matching the desired schema, then treat the generated output as your final \n",
"result.\n",
"\n",
":::note\n",
"\n",
"If you only need formatted values, try the [.with_structured_output()](/docs/how_to/structured_output/#the-with_structured_output-method) chat model method as a simpler entrypoint.\n",
"[Mistral](https://mistral.ai/), [OpenAI](https://openai.com/), and others, \n",
"support variants of a tool calling feature.\n",
"\n",
"LangChain implements standard interfaces for defining tools, passing them to LLMs, \n",
"and representing tool calls. This guide and the other How-to pages in the Tool section will show you how to use tools with LangChain."
"LangChain implements standard interfaces for defining tools, passing them to LLMs, and representing tool calls.\n",
"This guide will cover how to bind tools to an LLM, then invoke the LLM to generate these arguments."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Passing tools to chat models\n",
"## Defining tool schemas\n",
"\n",
"Chat models that support tool calling features implement a `.bind_tools` method, which \n",
"receives a list of LangChain [tool objects](https://api.python.langchain.com/en/latest/tools/langchain_core.tools.BaseTool.html#langchain_core.tools.BaseTool) \n",
"and binds them to the chat model in its expected format. Subsequent invocations of the \n",
"chat model will include tool schemas in its calls to the LLM.\n",
"For a model to be able to call tools, we need to pass in tool schemas that describe what the tool does and what it's arguments are. Chat models that support tool calling features implement a `.bind_tools()` method for passing tool schemas to the model. Tool schemas can be passed in as Python functions (with typehints and docstrings), Pydantic models, TypedDict classes, or LangChain [Tool objects](https://api.python.langchain.com/en/latest/tools/langchain_core.tools.BaseTool.html#langchain_core.tools.BaseTool). Subsequent invocations of the model will pass in these tool schemas along with the prompt.\n",
"\n",
"For example, we can define the schema for custom tools using the `@tool` decorator \n",
"on Python functions:"
"### Python functions\n",
"Our tool schemas can be Python functions:"
]
},
{
@@ -96,29 +67,41 @@
"metadata": {},
"outputs": [],
"source": [
"from langchain_core.tools import tool\n",
"\n",
"\n",
"@tool\n",
"# The function name, type hints, and docstring are all part of the tool\n",
"# schema that's passed to the model. Defining good, descriptive schemas\n",
"# is an extension of prompt engineering and is an important part of\n",
"# getting models to perform well.\n",
"def add(a: int, b: int) -> int:\n",
" \"\"\"Adds a and b.\"\"\"\n",
" \"\"\"Add two integers.\n",
"\n",
" Args:\n",
" a: First integer\n",
" b: Second integer\n",
" \"\"\"\n",
" return a + b\n",
"\n",
"\n",
"@tool\n",
"def multiply(a: int, b: int) -> int:\n",
" \"\"\"Multiplies a and b.\"\"\"\n",
" return a * b\n",
" \"\"\"Multiply two integers.\n",
"\n",
"\n",
"tools = [add, multiply]"
" Args:\n",
" a: First integer\n",
" b: Second integer\n",
" \"\"\"\n",
" return a * b"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Or below, we define the schema using [Pydantic](https://docs.pydantic.dev):"
"### LangChain Tool\n",
"\n",
"LangChain also implements a `@tool` decorator that allows for further control of the tool schema, such as tool names and argument descriptions. See the how-to guide [here](/docs/how_to/custom_tools/#creating-tools-from-functions) for details.\n",
"\n",
"### Pydantic class\n",
"\n",
"You can equivalently define the schemas without the accompanying functions using [Pydantic](https://docs.pydantic.dev):"
"As we can see, even though the prompt didn't really suggest a tool call, our LLM made one since it was forced to do so. You can look at the docs for [bind_tools()](https://api.python.langchain.com/en/latest/chat_models/langchain_openai.chat_models.base.BaseChatOpenAI.html#langchain_openai.chat_models.base.BaseChatOpenAI.bind_tools) to learn about all the ways to customize how your LLM selects tools."
"As we can see our LLM generated arguments to a tool! You can look at the docs for [bind_tools()](https://api.python.langchain.com/en/latest/chat_models/langchain_openai.chat_models.base.BaseChatOpenAI.html#langchain_openai.chat_models.base.BaseChatOpenAI.bind_tools) to learn about all the ways to customize how your LLM selects tools, as well as [this guide on how to force the LLM to call a tool](/docs/how_to/tool_choice/) rather than letting it decide."
]
},
{
@@ -242,21 +256,23 @@
},
{
"cell_type": "code",
"execution_count": 5,
"execution_count": 6,
"metadata": {},
"outputs": [
{
"data": {
"text/plain": [
"[{'name': 'Multiply',\n",
"[{'name': 'multiply',\n",
" 'args': {'a': 3, 'b': 12},\n",
" 'id': 'call_TnadLbWJu9HwDULRb51RNSMw'},\n",
" {'name': 'Add',\n",
" 'id': 'call_rcdMie7E89Xx06lEKKxJyB5N',\n",
" 'type': 'tool_call'},\n",
" {'name': 'add',\n",
" 'args': {'a': 11, 'b': 49},\n",
" 'id': 'call_Q9vt1up05sOQScXvUYWzSpCg'}]"
" 'id': 'call_nheGN8yfvSJsnIuGZaXihou3',\n",
" 'type': 'tool_call'}]"
]
},
"execution_count": 5,
"execution_count": 6,
"metadata": {},
"output_type": "execute_result"
}
@@ -278,31 +294,49 @@
"are populated in the `.invalid_tool_calls` attribute. An `InvalidToolCall` can have \n",
"a name, string arguments, identifier, and error message.\n",
"\n",
"If desired, [output parsers](/docs/how_to#output-parsers) can further \n",
"process the output. For example, we can convert existing values populated on the `.tool_calls` attribute back to the original Pydantic class using the\n",
"\n",
"## Parsing\n",
"\n",
"If desired, [output parsers](/docs/how_to#output-parsers) can further process the output. For example, we can convert existing values populated on the `.tool_calls` to Pydantic objects using the\n",
"This API is currently only supported by OpenAI.\n",
"\n",
":::\n",
"\n",
"OpenAI tool calling performs tool calling in parallel by default. That means that if we ask a question like \"What is the weather in Tokyo, New York, and Chicago?\" and we have a tool for getting the weather, it will call the tool 3 times in parallel. We can force it to call only a single tool once by using the ``parallel_tool_call`` parameter."
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
