In collaboration with @rlouf I build an
[outlines](https://dottxt-ai.github.io/outlines/latest/) integration for
langchain!
I think this is really useful for doing any type of structured output
locally.
[Dottxt](https://dottxt.co) spend alot of work optimising this process
at a lower level
([outlines-core](https://pypi.org/project/outlines-core/0.1.14/) written
in rust) so I think this is a better alternative over all current
approaches in langchain to do structured output.
It also implements the `.with_structured_output` method so it should be
a drop in replacement for a lot of applications.
The integration includes:
- **Outlines LLM class**
- **ChatOutlines class**
- **Tutorial Cookbooks**
- **Documentation Page**
- **Validation and error messages**
- **Exposes Outlines Structured output features**
- **Support for multiple backends**
- **Integration and Unit Tests**
Dependencies: `outlines` + additional (depending on backend used)
I am not sure if the unit-tests comply with all requirements, if not I
suggest to just remove them since I don't see a useful way to do it
differently.
### Quick overview:
Chat Models:
<img width="698" alt="image"
src="https://github.com/user-attachments/assets/05a499b9-858c-4397-a9ff-165c2b3e7acc">
Structured Output:
<img width="955" alt="image"
src="https://github.com/user-attachments/assets/b9fcac11-d3e5-4698-b1ae-8c4cb3d54c45">
---------
Co-authored-by: Vadym Barda <vadym@langchain.dev>
- **Description:** We have released the
[langchain-gigachat](https://github.com/ai-forever/langchain-gigachat?tab=readme-ov-file)
with new GigaChat integration that support's function/tool calling. This
PR deprecated legacy GigaChat class in community package.
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
Thank you for reading my first PR!
**Description:**
Deduplicate content in AzureSearch vectorstore.
Currently, by default, the content of the retrieval is placed both in
metadata and page_content of a Document.
This PR removes the content from metadata, and leaves it in
page_content.
**Issue:**:
Previously, the content was popped from result before metadata was
populated.
In #25828 , the order was changed which leads to a response with
duplicated content.
This was not the intention of that PR and seems undesirable.
Looking forward to seeing my contribution in the next version!
Cheers,
Renzo
**Description:** Add tool calling and structured output support for
SambaNovaCloud chat models, docs included
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
This change refines the handling of _model_kwargs in POST requests.
Instead of nesting _model_kwargs as a dictionary under the parameters
key, it is now directly unpacked and merged into the request's JSON
payload. This ensures that the model parameters are passed correctly and
avoids unnecessary nesting.E. g.:
```python
import asyncio
from langchain_huggingface.embeddings import HuggingFaceEndpointEmbeddings
embedding_input = ["This input will get multiplied" * 10000]
embeddings = HuggingFaceEndpointEmbeddings(
model="http://127.0.0.1:8081/embed",
model_kwargs={"truncate": True},
)
# Truncated parameters in synchronized methods are handled correctly
embeddings.embed_documents(texts=embedding_input)
# The truncate parameter is not handled correctly in the asynchronous method,
# and 413 Request Entity Too Large is returned.
asyncio.run(embeddings.aembed_documents(texts=embedding_input))
```
Co-authored-by: af su <saf@zjuici.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
Description:
* I'm planning to add `Document.id` support to the Chroma VectorStore,
but first I wanted to make sure all the integration tests were passing
first. They weren't. This PR fixes the broken tests.
* I found 2 issues:
* This change (from a year ago, exactly :) ) for supporting multi-modal
embeddings:
https://docs.trychroma.com/deployment/migration#migration-to-0.4.16---november-7,-2023
* This change https://github.com/langchain-ai/langchain/pull/27827 due
to an update in the chroma client.
Also ran `format` and `lint` on the changes.
Note: I am not a member of the Chroma team.
# Description
- adding stopReason to response_metadata to call stream and astream
- excluding NCP_APIGW_API_KEY input required validation
- to remove warning Field "model_name" has conflict with protected
namespace "model_".
cc. @vbarda
Description:
* Updated the OpenSearchVectorStore to use the `engine` parameter
captured at `init()` time as the default when adding documents to the
store.
Formatted, Linted, and Tested.
**Description:** The issue concerns the unexpected behavior observed
using the bind_tools method in LangChain's ChatOllama. When tools are
not bound, the llm.stream() method works as expected, returning
incremental chunks of content, which is crucial for real-time
applications such as conversational agents and live feedback systems.
However, when bind_tools([]) is used, the streaming behavior changes,
causing the output to be delivered in full chunks rather than
incrementally. This change negatively impacts the user experience by
breaking the real-time nature of the streaming mechanism.
**Issue:** #26971
---------
Co-authored-by: 4meyDam1e <amey.damle@mail.utoronto.ca>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Last week Anthropic released version 0.39.0 of its python sdk, which
enabled support for Python 3.13. This release deleted a legacy
`client.count_tokens` method, which we currently access during init of
the `Anthropic` LLM. Anthropic has replaced this functionality with the
[client.beta.messages.count_tokens()
API](https://github.com/anthropics/anthropic-sdk-python/pull/726).
To enable support for `anthropic >= 0.39.0` and Python 3.13, here we
drop support for the legacy token counting method, and add support for
the new method via `ChatAnthropic.get_num_tokens_from_messages`.
To fully support the token counting API, we update the signature of
`get_num_tokens_from_message` to accept tools everywhere.
---------
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Description:
* When working with OpenSearchVectorSearch to make
OpenSearchGraphVectorStore (coming soon), I noticed that there wasn't
type hinting for the underlying OpenSearch clients. This fixes that
issue.
* Confirmed tests are still passing with code changes.
Note that there is some additional code duplication now, but I think
this approach is cleaner overall.
## Description
As proposed in our earlier discussion #26977 we have introduced a Google
Books API Tool that leverages the Google Books API found at
[https://developers.google.com/books/docs/v1/using](https://developers.google.com/books/docs/v1/using)
to generate book recommendations.
### Sample Usage
```python
from langchain_community.tools import GoogleBooksQueryRun
from langchain_community.utilities import GoogleBooksAPIWrapper
api_wrapper = GoogleBooksAPIWrapper()
tool = GoogleBooksQueryRun(api_wrapper=api_wrapper)
tool.run('ai')
```
### Sample Output
```txt
Here are 5 suggestions based off your search for books related to ai:
1. "AI's Take on the Stigma Against AI-Generated Content" by Sandy Y. Greenleaf: In a world where artificial intelligence (AI) is rapidly advancing and transforming various industries, a new form of content creation has emerged: AI-generated content. However, despite its potential to revolutionize the way we produce and consume information, AI-generated content often faces a significant stigma. "AI's Take on the Stigma Against AI-Generated Content" is a groundbreaking book that delves into the heart of this issue, exploring the reasons behind the stigma and offering a fresh, unbiased perspective on the topic. Written from the unique viewpoint of an AI, this book provides readers with a comprehensive understanding of the challenges and opportunities surrounding AI-generated content. Through engaging narratives, thought-provoking insights, and real-world examples, this book challenges readers to reconsider their preconceptions about AI-generated content. It explores the potential benefits of embracing this technology, such as increased efficiency, creativity, and accessibility, while also addressing the concerns and drawbacks that contribute to the stigma. As you journey through the pages of this book, you'll gain a deeper understanding of the complex relationship between humans and AI in the realm of content creation. You'll discover how AI can be used as a tool to enhance human creativity, rather than replace it, and how collaboration between humans and machines can lead to unprecedented levels of innovation. Whether you're a content creator, marketer, business owner, or simply someone curious about the future of AI and its impact on our society, "AI's Take on the Stigma Against AI-Generated Content" is an essential read. With its engaging writing style, well-researched insights, and practical strategies for navigating this new landscape, this book will leave you equipped with the knowledge and tools needed to embrace the AI revolution and harness its potential for success. Prepare to have your assumptions challenged, your mind expanded, and your perspective on AI-generated content forever changed. Get ready to embark on a captivating journey that will redefine the way you think about the future of content creation.
Read more at https://play.google.com/store/books/details?id=4iH-EAAAQBAJ&source=gbs_api
2. "AI Strategies For Web Development" by Anderson Soares Furtado Oliveira: From fundamental to advanced strategies, unlock useful insights for creating innovative, user-centric websites while navigating the evolving landscape of AI ethics and security Key Features Explore AI's role in web development, from shaping projects to architecting solutions Master advanced AI strategies to build cutting-edge applications Anticipate future trends by exploring next-gen development environments, emerging interfaces, and security considerations in AI web development Purchase of the print or Kindle book includes a free PDF eBook Book Description If you're a web developer looking to leverage the power of AI in your projects, then this book is for you. Written by an AI and ML expert with more than 15 years of experience, AI Strategies for Web Development takes you on a transformative journey through the dynamic intersection of AI and web development, offering a hands-on learning experience.The first part of the book focuses on uncovering the profound impact of AI on web projects, exploring fundamental concepts, and navigating popular frameworks and tools. As you progress, you'll learn how to build smart AI applications with design intelligence, personalized user journeys, and coding assistants. Later, you'll explore how to future-proof your web development projects using advanced AI strategies and understand AI's impact on jobs. Toward the end, you'll immerse yourself in AI-augmented development, crafting intelligent web applications and navigating the ethical landscape.Packed with insights into next-gen development environments, AI-augmented practices, emerging realities, interfaces, and security governance, this web development book acts as your roadmap to staying ahead in the AI and web development domain. What you will learn Build AI-powered web projects with optimized models Personalize UX dynamically with AI, NLP, chatbots, and recommendations Explore AI coding assistants and other tools for advanced web development Craft data-driven, personalized experiences using pattern recognition Architect effective AI solutions while exploring the future of web development Build secure and ethical AI applications following TRiSM best practices Explore cutting-edge AI and web development trends Who this book is for This book is for web developers with experience in programming languages and an interest in keeping up with the latest trends in AI-powered web development. Full-stack, front-end, and back-end developers, UI/UX designers, software engineers, and web development enthusiasts will also find valuable information and practical guidelines for developing smarter websites with AI. To get the most out of this book, it is recommended that you have basic knowledge of programming languages such as HTML, CSS, and JavaScript, as well as a familiarity with machine learning concepts.
Read more at https://play.google.com/store/books/details?id=FzYZEQAAQBAJ&source=gbs_api
3. "Artificial Intelligence for Students" by Vibha Pandey: A multifaceted approach to develop an understanding of AI and its potential applications KEY FEATURES ● AI-informed focuses on AI foundation, applications, and methodologies. ● AI-inquired focuses on computational thinking and bias awareness. ● AI-innovate focuses on creative and critical thinking and the Capstone project. DESCRIPTION AI is a discipline in Computer Science that focuses on developing intelligent machines, machines that can learn and then teach themselves. If you are interested in AI, this book can definitely help you prepare for future careers in AI and related fields. The book is aligned with the CBSE course, which focuses on developing employability and vocational competencies of students in skill subjects. The book is an introduction to the basics of AI. It is divided into three parts – AI-informed, AI-inquired and AI-innovate. It will help you understand AI's implications on society and the world. You will also develop a deeper understanding of how it works and how it can be used to solve complex real-world problems. Additionally, the book will also focus on important skills such as problem scoping, goal setting, data analysis, and visualization, which are essential for success in AI projects. Lastly, you will learn how decision trees, neural networks, and other AI concepts are commonly used in real-world applications. By the end of the book, you will develop the skills and competencies required to pursue a career in AI. WHAT YOU WILL LEARN ● Get familiar with the basics of AI and Machine Learning. ● Understand how and where AI can be applied. ● Explore different applications of mathematical methods in AI. ● Get tips for improving your skills in Data Storytelling. ● Understand what is AI bias and how it can affect human rights. WHO THIS BOOK IS FOR This book is for CBSE class XI and XII students who want to learn and explore more about AI. Basic knowledge of Statistical concepts, Algebra, and Plotting of equations is a must. TABLE OF CONTENTS 1. Introduction: AI for Everyone 2. AI Applications and Methodologies 3. Mathematics in Artificial Intelligence 4. AI Values (Ethical Decision-Making) 5. Introduction to Storytelling 6. Critical and Creative Thinking 7. Data Analysis 8. Regression 9. Classification and Clustering 10. AI Values (Bias Awareness) 11. Capstone Project 12. Model Lifecycle (Knowledge) 13. Storytelling Through Data 14. AI Applications in Use in Real-World
Read more at https://play.google.com/store/books/details?id=ptq1EAAAQBAJ&source=gbs_api
4. "The AI Book" by Ivana Bartoletti, Anne Leslie and Shân M. Millie: Written by prominent thought leaders in the global fintech space, The AI Book aggregates diverse expertise into a single, informative volume and explains what artifical intelligence really means and how it can be used across financial services today. Key industry developments are explained in detail, and critical insights from cutting-edge practitioners offer first-hand information and lessons learned. Coverage includes: · Understanding the AI Portfolio: from machine learning to chatbots, to natural language processing (NLP); a deep dive into the Machine Intelligence Landscape; essentials on core technologies, rethinking enterprise, rethinking industries, rethinking humans; quantum computing and next-generation AI · AI experimentation and embedded usage, and the change in business model, value proposition, organisation, customer and co-worker experiences in today’s Financial Services Industry · The future state of financial services and capital markets – what’s next for the real-world implementation of AITech? · The innovating customer – users are not waiting for the financial services industry to work out how AI can re-shape their sector, profitability and competitiveness · Boardroom issues created and magnified by AI trends, including conduct, regulation & oversight in an algo-driven world, cybersecurity, diversity & inclusion, data privacy, the ‘unbundled corporation’ & the future of work, social responsibility, sustainability, and the new leadership imperatives · Ethical considerations of deploying Al solutions and why explainable Al is so important
Read more at http://books.google.ca/books?id=oE3YDwAAQBAJ&dq=ai&hl=&source=gbs_api
5. "Artificial Intelligence in Society" by OECD: The artificial intelligence (AI) landscape has evolved significantly from 1950 when Alan Turing first posed the question of whether machines can think. Today, AI is transforming societies and economies. It promises to generate productivity gains, improve well-being and help address global challenges, such as climate change, resource scarcity and health crises.
Read more at https://play.google.com/store/books/details?id=eRmdDwAAQBAJ&source=gbs_api
```
## Issue
This closes#27276
## Dependencies
No additional dependencies were added
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
all template installs will now have to declare `--branch v0.2` to make
clear they aren't compatible with langchain 0.3 (most have a pydantic v1
setup). e.g.
```
langchain-cli app add pirate-speak --branch v0.2
```
Thank you for contributing to LangChain!
- [x] **PR title**: "community: chat models wrapper for Cloudflare
Workers AI"
- [x] **PR message**:
- **Description:** Add chat models wrapper for Cloudflare Workers AI.
Enables Langgraph intergration via ChatModel for tool usage, agentic
usage.
- [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.
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
- **Description:** This PR adds functionality to pass in in-memory bytes
as a source to `AzureAIDocumentIntelligenceLoader`.
- **Issue:** I needed the functionality, so I added it.
- **Dependencies:** NA
- **Twitter handle:** @akseljoonas if this is a big enough change :)
---------
Co-authored-by: Aksel Joonas Reedi <aksel@klippa.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
# OCR-based PDF loader
This implements [Zerox](https://github.com/getomni-ai/zerox) PDF
document loader.
Zerox utilizes simple but very powerful (even though slower and more
costly) approach to parsing PDF documents: it converts PDF to series of
images and passes it to a vision model requesting the contents in
markdown.
It is especially suitable for complex PDFs that are not parsed well by
other alternatives.
## Example use:
```python
from langchain_community.document_loaders.pdf import ZeroxPDFLoader
os.environ["OPENAI_API_KEY"] = "" ## your-api-key
model = "gpt-4o-mini" ## openai model
pdf_url = "https://assets.ctfassets.net/f1df9zr7wr1a/soP1fjvG1Wu66HJhu3FBS/034d6ca48edb119ae77dec5ce01a8612/OpenAI_Sacra_Teardown.pdf"
loader = ZeroxPDFLoader(file_path=pdf_url, model=model)
docs = loader.load()
```
The Zerox library supports wide range of provides/models. See Zerox
documentation for details.
- **Dependencies:** `zerox`
- **Twitter handle:** @martintriska1
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
---------
Co-authored-by: Erick Friis <erickfriis@gmail.com>
## Description
This PR adds support for Memcached as a usable LLM model cache by adding
the ```MemcachedCache``` implementation relying on the
[pymemcache](https://github.com/pinterest/pymemcache) client.
Unit test-wise, the new integration is generally covered under existing
import testing. All new functionality depends on pymemcache if
instantiated and used, so to comply with the other cache implementations
the PR also adds optional integration tests for ```MemcachedCache```.
Since this is a new integration, documentation is added for Memcached as
an integration and as an LLM Cache.
## Issue
This PR closes#27275 which was originally raised as a discussion in
#27035
## Dependencies
There are no new required dependencies for langchain, but
[pymemcache](https://github.com/pinterest/pymemcache) is required to
instantiate the new ```MemcachedCache```.
## Example Usage
```python3
from langchain.globals import set_llm_cache
from langchain_openai import OpenAI
from langchain_community.cache import MemcachedCache
from pymemcache.client.base import Client
llm = OpenAI(model="gpt-3.5-turbo-instruct", n=2, best_of=2)
set_llm_cache(MemcachedCache(Client('localhost')))
# The first time, it is not yet in cache, so it should take longer
llm.invoke("Which city is the most crowded city in the USA?")
# The second time it is, so it goes faster
llm.invoke("Which city is the most crowded city in the USA?")
```
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
## What this PR does?
### Currently `O365BaseLoader` (and consequently both derived loaders)
are limited to `pdf`, `doc`, `docx` files.
- **Solution: here we introduce _handlers_ attribute that allows for
custom handlers to be passed in. This is done in _dict_ form:**
**Example:**
```python
from langchain_community.document_loaders.parsers.documentloader_adapter import DocumentLoaderAsParser
# PR for DocumentLoaderAsParser here: https://github.com/langchain-ai/langchain/pull/27749
from langchain_community.document_loaders.excel import UnstructuredExcelLoader
xlsx_parser = DocumentLoaderAsParser(UnstructuredExcelLoader, mode="paged")
# create dictionary mapping file types to handlers (parsers)
handlers = {
"doc": MsWordParser()
"pdf": PDFMinerParser()
"txt": TextParser()
"xlsx": xlsx_parser
}
loader = SharePointLoader(document_library_id="...",
handlers=handlers # pass handlers to SharePointLoader
)
documents = loader.load()
# works the same in OneDriveLoader
loader = OneDriveLoader(document_library_id="...",
handlers=handlers
)
```
This dictionary is then passed to `MimeTypeBasedParser` same as in the
[current
implementation](5a2cfb49e0/libs/community/langchain_community/document_loaders/parsers/registry.py (L13)).
### Currently `SharePointLoader` and `OneDriveLoader` are separate
loaders that both inherit from `O365BaseLoader`
However both of these implement the same functionality. The only
differences are:
- `SharePointLoader` requires argument `document_library_id` whereas
`OneDriveLoader` requires `drive_id`. These are just different names for
the same thing.
- `SharePointLoader` implements significantly more features.
- **Solution: `OneDriveLoader` is replaced with an empty shell just
renaming `drive_id` to `document_library_id` and inheriting from
`SharePointLoader`**
**Dependencies:** None
**Twitter handle:** @martintriska1
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, etc. is
being modified. Use "docs: ..." for purely docs changes, "templates:
..." for template changes, "infra: ..." for CI changes.
- Example: "community: add foobar LLM"
- [ ] **description**
langchain_core.runnables.graph_mermaid.draw_mermaid_png calls this
function, but the Mermaid API returns JPEG by default. To be consistent,
add the option `file_type` with the default `png` type.
- [ ] **Add tests and docs**: If you're adding a new integration, please
include
With this small change, I didn't add tests and docs.
- [ ] **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:
One long sentence was divided into two.
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.
Now `encode_kwargs` used for both for documents and queries and this
leads to wrong embeddings. E. g.:
```python
model_kwargs = {"device": "cuda", "trust_remote_code": True}
encode_kwargs = {"normalize_embeddings": False, "prompt_name": "s2p_query"}
model = HuggingFaceEmbeddings(
model_name="dunzhang/stella_en_400M_v5",
model_kwargs=model_kwargs,
encode_kwargs=encode_kwargs,
)
query_embedding = np.array(
model.embed_query("What are some ways to reduce stress?",)
)
document_embedding = np.array(
model.embed_documents(
[
"There are many effective ways to reduce stress. Some common techniques include deep breathing, meditation, and physical activity. Engaging in hobbies, spending time in nature, and connecting with loved ones can also help alleviate stress. Additionally, setting boundaries, practicing self-care, and learning to say no can prevent stress from building up.",
"Green tea has been consumed for centuries and is known for its potential health benefits. It contains antioxidants that may help protect the body against damage caused by free radicals. Regular consumption of green tea has been associated with improved heart health, enhanced cognitive function, and a reduced risk of certain types of cancer. The polyphenols in green tea may also have anti-inflammatory and weight loss properties.",
]
)
)
print(model._client.similarity(query_embedding, document_embedding)) # output: tensor([[0.8421, 0.3317]], dtype=torch.float64)
```
But from the [model
card](https://huggingface.co/dunzhang/stella_en_400M_v5#sentence-transformers)
expexted like this:
```python
model_kwargs = {"device": "cuda", "trust_remote_code": True}
encode_kwargs = {"normalize_embeddings": False}
query_encode_kwargs = {"normalize_embeddings": False, "prompt_name": "s2p_query"}
model = HuggingFaceEmbeddings(
model_name="dunzhang/stella_en_400M_v5",
model_kwargs=model_kwargs,
encode_kwargs=encode_kwargs,
query_encode_kwargs=query_encode_kwargs,
)
query_embedding = np.array(
model.embed_query("What are some ways to reduce stress?", )
)
document_embedding = np.array(
model.embed_documents(
[
"There are many effective ways to reduce stress. Some common techniques include deep breathing, meditation, and physical activity. Engaging in hobbies, spending time in nature, and connecting with loved ones can also help alleviate stress. Additionally, setting boundaries, practicing self-care, and learning to say no can prevent stress from building up.",
"Green tea has been consumed for centuries and is known for its potential health benefits. It contains antioxidants that may help protect the body against damage caused by free radicals. Regular consumption of green tea has been associated with improved heart health, enhanced cognitive function, and a reduced risk of certain types of cancer. The polyphenols in green tea may also have anti-inflammatory and weight loss properties.",
]
)
)
print(model._client.similarity(query_embedding, document_embedding)) # tensor([[0.8398, 0.2990]], dtype=torch.float64)
```
There was a change of attribute name which was "max_batch_size". It's
now "get_max_batch_size" method.
I want to use "create_batches" which is right down below.
Please check this PR link.
reference: https://github.com/chroma-core/chroma/pull/2305
---------
Signed-off-by: Prithvi Kannan <prithvi.kannan@databricks.com>
Co-authored-by: Prithvi Kannan <46332835+prithvikannan@users.noreply.github.com>
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
Co-authored-by: Jun Yamog <jkyamog@gmail.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Co-authored-by: ono-hiroki <86904208+ono-hiroki@users.noreply.github.com>
Co-authored-by: Dobiichi-Origami <56953648+Dobiichi-Origami@users.noreply.github.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Co-authored-by: Duy Huynh <vndee.huynh@gmail.com>
Co-authored-by: Rashmi Pawar <168514198+raspawar@users.noreply.github.com>
Co-authored-by: sifatj <26035630+sifatj@users.noreply.github.com>
Co-authored-by: Eric Pinzur <2641606+epinzur@users.noreply.github.com>
Co-authored-by: Daniel Vu Dao <danielvdao@users.noreply.github.com>
Co-authored-by: Ofer Mendelevitch <ofermend@gmail.com>
Co-authored-by: Stéphane Philippart <wildagsx@gmail.com>
- **Description:** change to do the batch embedding server side and not
client side
- **Twitter handle:** @wildagsx
---------
Co-authored-by: ccurme <chester.curme@gmail.com>
Description:
This fixes an issue that mistakenly created in
https://github.com/langchain-ai/langchain/pull/27253. The issue
currently exists only in `langchain-community==0.3.4`.
Test cases were added to prevent this issue in the future.
Co-authored-by: Erick Friis <erick@langchain.dev>
### Description:
This PR sets a default value of `output_token_limit = 4000` for the
`PowerBIToolkit` to fix the unintentionally validation error.
### Problem:
When attempting to run a code snippet from [Langchain's PowerBI toolkit
documentation](https://python.langchain.com/v0.1/docs/integrations/toolkits/powerbi/)
to interact with a `PowerBIDataset`, the following error occurs:
```
pydantic.v1.error_wrappers.ValidationError: 1 validation error for QueryPowerBITool
output_token_limit
none is not an allowed value (type=type_error.none.not_allowed)
```
### Root Cause:
The issue arises because when creating a `QueryPowerBITool`, the
`output_token_limit` parameter is unintentionally set to `None`, which
is the current default for `PowerBIToolkit`. However, `QueryPowerBITool`
expects a default value of `4000` for `output_token_limit`. This
unintended override causes the error.
17659ca2cd/libs/community/langchain_community/agent_toolkits/powerbi/toolkit.py (L63)17659ca2cd/libs/community/langchain_community/agent_toolkits/powerbi/toolkit.py (L72-L79)17659ca2cd/libs/community/langchain_community/tools/powerbi/tool.py (L39)
### Solution:
To resolve this, the default value of `output_token_limit` is now
explicitly set to `4000` in `PowerBIToolkit` to prevent the accidental
assignment of `None`.
Co-authored-by: ccurme <chester.curme@gmail.com>
- **Description:**
Currently CommaSeparatedListOutputParser can't handle strings that may
contain commas within a column. It would parse any commas as the
delimiter.
Ex.
"foo, foo2", "bar", "baz"
It will create 4 columns: "foo", "foo2", "bar", "baz"
This should be 3 columns:
"foo, foo2", "bar", "baz"
- **Dependencies:**
Added 2 additional imports, but they are built in python packages.
import csv
from io import StringIO
- **Twitter handle:** @jkyamog
- [ ] **Add tests and docs**:
1. added simple unit test test_multiple_items_with_comma
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
Co-authored-by: Bagatur <22008038+baskaryan@users.noreply.github.com>
Co-authored-by: Bagatur <baskaryan@gmail.com>
Thank you for contributing to LangChain!
- [x] **PR title**: "core: use friendlier names for duplicated nodes in
mermaid output"
- **Description:** When generating the Mermaid visualization of a chain,
if the chain had multiple nodes of the same type, the reid function
would replace their names with the UUID node_id. This made the generated
graph difficult to understand. This change deduplicates the nodes in a
chain by appending an index to their names.
- **Issue:** None
- **Discussion:**
https://github.com/langchain-ai/langchain/discussions/27714
- **Dependencies:** None
- [ ] **Add tests and docs**:
- Currently this functionality is not covered by unit tests, happy to
add tests if you'd like
- [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.
# Example Code:
```python
from langchain_core.runnables import RunnablePassthrough
def fake_llm(prompt: str) -> str: # Fake LLM for the example
return "completion"
runnable = {
'llm1': fake_llm,
'llm2': fake_llm,
} | RunnablePassthrough.assign(
total_chars=lambda inputs: len(inputs['llm1'] + inputs['llm2'])
)
print(runnable.get_graph().draw_mermaid(with_styles=False))
```
# Before
```mermaid
graph TD;
Parallel_llm1_llm2_Input --> 0b01139db5ed4587ad37964e3a40c0ec;
0b01139db5ed4587ad37964e3a40c0ec --> Parallel_llm1_llm2_Output;
Parallel_llm1_llm2_Input --> a98d4b56bd294156a651230b9293347f;
a98d4b56bd294156a651230b9293347f --> Parallel_llm1_llm2_Output;
Parallel_total_chars_Input --> Lambda;
Lambda --> Parallel_total_chars_Output;
Parallel_total_chars_Input --> Passthrough;
Passthrough --> Parallel_total_chars_Output;
Parallel_llm1_llm2_Output --> Parallel_total_chars_Input;
```
# After
```mermaid
graph TD;
Parallel_llm1_llm2_Input --> fake_llm_1;
fake_llm_1 --> Parallel_llm1_llm2_Output;
Parallel_llm1_llm2_Input --> fake_llm_2;
fake_llm_2 --> Parallel_llm1_llm2_Output;
Parallel_total_chars_Input --> Lambda;
Lambda --> Parallel_total_chars_Output;
Parallel_total_chars_Input --> Passthrough;
Passthrough --> Parallel_total_chars_Output;
Parallel_llm1_llm2_Output --> Parallel_total_chars_Input;
```
PR title: “langchain: add batch request support for text-embedding-v3
model”
PR message:
• Description: This PR introduces batch request support for the
text-embedding-v3 model within LangChain. The new functionality allows
users to process multiple text inputs in a single request, improving
efficiency and performance for high-volume applications.
• Issue: This PR addresses #<issue_number> (if applicable).
• Dependencies: No new external dependencies are required for this
change.
• Twitter handle: If announced on Twitter, please mention me at
@yourhandle.
Add tests and docs:
1. Added unit tests to cover the batch request functionality, ensuring
it operates without requiring network access.
2. Included an example notebook demonstrating the batch request feature,
located in docs/docs/integrations.
Lint and test: All required formatting and linting checks have been
performed using make format and make lint. The changes have been
verified with make test to ensure compatibility.
Additional notes:
• The changes are fully backwards compatible.
• No modifications were made to pyproject.toml, ensuring no new
dependencies were added.
• The update only affects the langchain package and does not involve
other packages.
---------
Co-authored-by: Chester Curme <chester.curme@gmail.com>
…ING} {code: Neo.ClientNotification.Statement.FeatureDeprecationWarning}
{category: DEPRECATION} {title: This feature is deprecated and will be
removed in future versions.} {description: CALL subquery without a
variable scope clause is now deprecated." this warning
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core, 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.
Co-authored-by: putao520 <putao520@putao282.com>
The metadata["source"] value for the web paths was being set to
temporary path (/tmp).
Fixed it by creating a new variable self.original_file_path, which will
store the original path.
Thank you for contributing to LangChain!
- [ ] **PR title**: "package: description"
- Where "package" is whichever of langchain, community, core, 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.
I will keep this PR as small as the changes made.
**Description:** fixes a fatal bug syntax error in
AzureCosmosDBNoSqlVectorSearch
**Issue:** #27269#25468
**Description:**
Update the wrapper to support the Polygon API if not you get an error. I
keeped `STOCKBUSINESS` for retro-compatbility with older endpoints /
other uses
Old Code:
```
if status not in ("OK", "STOCKBUSINESS"):
raise ValueError(f"API Error: {data}")
```
API Respond:
```
API Error: {'results': {'P': 0.22, 'S': 0, 'T': 'ZOM', 'X': 5, 'p': 0.123, 'q': 0, 's': 200, 't': 1729614422813395456, 'x': 1, 'z': 1}, 'status': 'STOCKSBUSINESS', 'request_id': 'XXXXXX'}
```
- **Issue:** N/A Polygon API update
- **Dependencies:** N/A
- **Twitter handle:** @wgcv
---------
Co-authored-by: ccurme <chester.curme@gmail.com>
- **Description:** add missing tool_calls kwargs of delta message in
openai adapter, then tool call will work correctly via adapter's stream
chat completion
- **Issue:** Fixes
https://github.com/langchain-ai/langchain/issues/25436
- **Dependencies:** None
- **Description:** Change `MoonshotCommon.client` type from
`_MoonshotClient` to `Any`.
- **Issue:** Fix the issue #27058
- **Dependencies:** No
- **Twitter handle:** TaoWang2218
In PR #17100, the implementation for Moonshot was added, which defined
two classes:
- `MoonshotChat(MoonshotCommon, ChatOpenAI)` in
`langchain_community.chat_models.moonshot`;
- Here, `validate_environment()` assigns **client** as
`openai.OpenAI().chat.completions`
- Note that **client** here is actually a member variable defined in
`ChatOpenAI`;
- `MoonshotCommon` in `langchain_community.llms.moonshot`;
- And here, `validate_environment()` assigns **_client** as
`_MoonshotClient`;
- Note that this is the underscored **_client**, which is defined within
`MoonshotCommon` itself;
At this time, there was no conflict between the two, one being `client`
and the other `_client`.
However, in PR #25878 which fixed#24390, `_client` in `MoonshotCommon`
was changed to `client`. Since then, a conflict in the definition of
`client` has arisen between `MoonshotCommon` and `MoonshotChat`, which
caused `pydantic` validation error.
To fix this issue, the type of `client` in `MoonshotCommon` should be
changed to `Any`.
Signed-off-by: Tao Wang <twang2218@gmail.com>
Thank you for contributing to LangChain!
- **Description:** Adding an empty metadata field when metadata is not
present in the data
- **Issue:** This PR fixes the issue when the data items doesn't contain
the metadata field. This happens when there is already data in the
container, or cx uses CosmosDB Python SDK to insert data.
- **Dependencies:** No dependencies required
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.
- [x] **PR title**:
"community: OCI Generative AI tool calling bug fix
- [x] **PR message**:
- **Description:** bug fix for streaming chat responses with tool calls.
Update to PR 24693
- **Issue:** chat response content is repeated when streaming
- **Dependencies:** NA
- **Twitter handle:** NA
- [x] **Add tests and docs**: NA
- [x] **Lint and test**: make format, make lint and make test we run
successfully
---------
Co-authored-by: Arthur Cheng <arthur.cheng@oracle.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
### About
- **Description:** In the Gitlab utilities used for the Gitlab tool
there is no check to prevent pushing to the main branch, as this is
already done for Github (for example here:
5a2cfb49e0/libs/community/langchain_community/utilities/github.py (L587)).
This PR add this check as already done for Github.
- **Issue:** None
- **Dependencies:** None
**Description:** Add support for Writer chat models
**Issue:** N/A
**Dependencies:** Add `writer-sdk` to optional dependencies.
**Twitter handle:** Please tag `@samjulien` and `@Get_Writer`
**Tests and docs**
- [x] Unit test
- [x] Example notebook in `docs/docs/integrations` directory.
**Lint and test**
- [x] Run `make format`
- [x] Run `make lint`
- [x] Run `make test`
---------
Co-authored-by: Johannes <tolstoy.work@gmail.com>
Co-authored-by: Erick Friis <erick@langchain.dev>
**Description:**
- Fix bug in Replicate LLM class, where it was looking for parameter
names in a place where they no longer exist in pydantic 2, resulting in
the "Field required" validation error described in the issue.
- Fix Replicate LLM integration tests to:
- Use active models on Replicate.
- Use the correct model parameter `max_new_tokens` as shown in the
[Replicate
docs](https://replicate.com/docs/guides/language-models/how-to-use#minimum-and-maximum-new-tokens).
- Use callbacks instead of deprecated callback_manager.
**Issue:** #26937
**Dependencies:** n/a
**Twitter handle:** n/a
---------
Signed-off-by: Fayvor Love <fayvor@gmail.com>
Co-authored-by: Chester Curme <chester.curme@gmail.com>
Thank you for contributing to LangChain!
**Description:** Added the model parameters to be passed in the OpenAI
Assistant. Enabled it at the `OpenAIAssistantV2Runnable` class.
**Issue:** NA
**Dependencies:** None
**Twitter handle:** luizf0992
Thank you for contributing to LangChain!
- **Description:** Add token_usage and model_name metadata to
ChatZhipuAI stream() and astream() response
- **Issue:** None
- **Dependencies:** None
- **Twitter handle:** None
- [ ] **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: jianfehuang <jianfehuang@tencent.com>
**Description:**
- Add the `lora_request` parameter to the VLLM class to support LoRA
model configurations. This enhancement allows users to specify LoRA
requests directly when using VLLM, enabling more flexible and efficient
model customization.
**Issue:**
- No existing issue for `lora_adapter` in VLLM. This PR addresses the
need for configuring LoRA requests within the VLLM framework.
- Reference : [Using LoRA Adapters in
vLLM](https://docs.vllm.ai/en/stable/models/lora.html#using-lora-adapters)
**Example Code :**
Before this change, the `lora_request` parameter was not applied
correctly:
```python
ADAPTER_PATH = "/path/of/lora_adapter"
llm = VLLM(model="Bllossom/llama-3.2-Korean-Bllossom-3B",
max_new_tokens=512,
top_k=2,
top_p=0.90,
temperature=0.1,
vllm_kwargs={
"gpu_memory_utilization":0.5,
"enable_lora":True,
"max_model_len":1024,
}
)
print(llm.invoke(
["...prompt_content..."],
lora_request=LoRARequest("lora_adapter", 1, ADAPTER_PATH)
))
```
**Before Change Output:**
```bash
response was not applied lora_request
```
So, I attempted to apply the lora_adapter to
langchain_community.llms.vllm.VLLM.
**current output:**
```bash
response applied lora_request
```
**Dependencies:**
- None
**Lint and test:**
- All tests and lint checks have passed.
---------
Co-authored-by: Um Changyong <changyong.um@sfa.co.kr>
**Description:**
This PR fixes an issue where non-ASCII characters in Pydantic field
descriptions were being escaped to their Unicode representations when
using `JsonOutputParser`. The change allows non-ASCII characters to be
preserved in the output, which is especially important for multilingual
support and when working with non-English languages.
**Issue:** Fixes#27256
**Example Code:**
```python
from pydantic import BaseModel, Field
from langchain_core.output_parsers import JsonOutputParser
class Article(BaseModel):
title: str = Field(description="科学文章的标题")
output_data_structure = Article
parser = JsonOutputParser(pydantic_object=output_data_structure)
print(parser.get_format_instructions())
```
**Previous Output**:
```... "title": {"description": "\\u79d1\\u5b66\\u6587\\u7ae0\\u7684\\u6807\\u9898", "title": "Title", "type": "string"}} ...```
**Current Output**:
```... "title": {"description": "科学文章的标题", "title": "Title", "type":
"string"}} ...```
**Changes made**:
- Modified `json.dumps()` call in
`langchain_core/output_parsers/json.py` to use `ensure_ascii=False`
- Added a unit test to verify Unicode handling
Co-authored-by: Harsimran-19 <harsimran1869@gmail.com>
**Description:**
When annotating a function with the @tool decorator, the symbol should
have type BaseTool. The previous type annotations did not convey that to
type checkers. This patch creates 4 overloads for the tool function for
the 4 different use cases.
1. @tool decorator with no arguments
2. @tool decorator with only keyword arguments
3. @tool decorator with a name argument (and possibly keyword arguments)
4. Invoking tool as function with a name and runnable positional
arguments
The main function is updated to match the overloads. The changes are
100% backwards compatible (all existing calls should continue to work,
just with better type annotations).
**Twitter handle:** @nvachhar
---------
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
**Description:** Fixes None addition issues when an empty value is
passed on
If no one reviews your PR within a few days, please @-mention one of
baskaryan, efriis, eyurtsev, ccurme, vbarda, hwchase17.
**Description:**
The `aiohttp.ClientSession` is closed at the end of the with statement,
which causes an error during a second call.
The implemented fix is to define the session directly within the with
block, exactly like in the textembed code:
c6350d636e/libs/community/langchain_community/embeddings/textembed.py (L335-L346)
**Issue:** Fix#26932
Co-authored-by: ccurme <chester.curme@gmail.com>
**Description:** We improve the performance of the InMemoryVectorStore.
**Isue:** Originally, similarity was computed document by document:
```
for doc in self.store.values():
vector = doc["vector"]
similarity = float(cosine_similarity([embedding], [vector]).item(0))
```
This is inefficient and does not make use of numpy vectorization.
This PR computes the similarity in one vectorized go:
```
docs = list(self.store.values())
similarity = cosine_similarity([embedding], [doc["vector"] for doc in docs])
```
**Dependencies:** None
**Twitter handle:** @b12_consulting, @Vincent_Min
---------
Co-authored-by: Eugene Yurtsev <eyurtsev@gmail.com>
**Description:** Returns the document id along with the Vector Search
results
**Issue:** Fixes https://github.com/langchain-ai/langchain/issues/26860
for CouchbaseVectorStore
- [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.
Co-authored-by: Erick Friis <erick@langchain.dev>
Reopened as a personal repo outside the organization.
## Description
- Naver HyperCLOVA X community package
- Add chat model & embeddings
- Add unit test & integration test
- Add chat model & embeddings docs
- I changed partner
package(https://github.com/langchain-ai/langchain/pull/24252) to
community package on this PR
- Could this
embeddings(https://github.com/langchain-ai/langchain/pull/21890) be
deprecated? We are trying to replace it with embedding
model(**ClovaXEmbeddings**) in this PR.
Twitter handle: None. (if needed, contact with
joonha.jeon@navercorp.com)
---
you can check our previous discussion below:
> one question on namespaces - would it make sense to have these in
.clova namespaces instead of .naver?
I would like to keep it as is, unless it is essential to unify the
package name.
(ClovaX is a branding for the model, and I plan to add other models and
components. They need to be managed as separate classes.)
> also, could you clarify the difference between ClovaEmbeddings and
ClovaXEmbeddings?
There are 3 models that are being serviced by embedding, and all are
supported in the current PR. In addition, all the functionality of CLOVA
Studio that serves actual models, such as distinguishing between test
apps and service apps, is supported. The existing PR does not support
this content because it is hard-coded.
---------
Co-authored-by: Erick Friis <erick@langchain.dev>
Co-authored-by: Vadym Barda <vadym@langchain.dev>
## Description
I encountered an error while using the` gemma-2-2b-it model` with the
`HuggingFacePipeline` class and have implemented a fix to resolve this
issue.
### What is Problem
```python
model_id="google/gemma-2-2b-it"
gemma_2_model = AutoModelForCausalLM.from_pretrained(model_id)
gemma_2_tokenizer = AutoTokenizer.from_pretrained(model_id)
gen = pipeline(
task='text-generation',
model=gemma_2_model,
tokenizer=gemma_2_tokenizer,
max_new_tokens=1024,
device=0 if torch.cuda.is_available() else -1,
temperature=.5,
top_p=0.7,
repetition_penalty=1.1,
do_sample=True,
)
llm = HuggingFacePipeline(pipeline=gen)
for chunk in llm.stream("Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World."):
print(chunk, end="", flush=True)
```
This code outputs the following error message:
```
/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py:1258: UserWarning: Using the model-agnostic default `max_length` (=20) to control the generation length. We recommend setting `max_new_tokens` to control the maximum length of the generation.
warnings.warn(
Exception in thread Thread-19 (generate):
Traceback (most recent call last):
File "/usr/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
self.run()
File "/usr/lib/python3.10/threading.py", line 953, in run
self._target(*self._args, **self._kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py", line 1874, in generate
self._validate_generated_length(generation_config, input_ids_length, has_default_max_length)
File "/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py", line 1266, in _validate_generated_length
raise ValueError(
ValueError: Input length of input_ids is 31, but `max_length` is set to 20. This can lead to unexpected behavior. You should consider increasing `max_length` or, better yet, setting `max_new_tokens`.
```
In addition, the following error occurs when the number of tokens is
reduced.
```python
for chunk in llm.stream("Hello World"):
print(chunk, end="", flush=True)
```
```
/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py:1258: UserWarning: Using the model-agnostic default `max_length` (=20) to control the generation length. We recommend setting `max_new_tokens` to control the maximum length of the generation.
warnings.warn(
/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py:1885: UserWarning: You are calling .generate() with the `input_ids` being on a device type different than your model's device. `input_ids` is on cpu, whereas the model is on cuda. You may experience unexpected behaviors or slower generation. Please make sure that you have put `input_ids` to the correct device by calling for example input_ids = input_ids.to('cuda') before running `.generate()`.
warnings.warn(
Exception in thread Thread-20 (generate):
Traceback (most recent call last):
File "/usr/lib/python3.10/threading.py", line 1016, in _bootstrap_inner
self.run()
File "/usr/lib/python3.10/threading.py", line 953, in run
self._target(*self._args, **self._kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/utils/_contextlib.py", line 116, in decorate_context
return func(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py", line 2024, in generate
result = self._sample(
File "/usr/local/lib/python3.10/dist-packages/transformers/generation/utils.py", line 2982, in _sample
outputs = self(**model_inputs, return_dict=True)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1562, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/transformers/models/gemma2/modeling_gemma2.py", line 994, in forward
outputs = self.model(
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1562, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/transformers/models/gemma2/modeling_gemma2.py", line 803, in forward
inputs_embeds = self.embed_tokens(input_ids)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1553, in _wrapped_call_impl
return self._call_impl(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/module.py", line 1562, in _call_impl
return forward_call(*args, **kwargs)
File "/usr/local/lib/python3.10/dist-packages/torch/nn/modules/sparse.py", line 164, in forward
return F.embedding(
File "/usr/local/lib/python3.10/dist-packages/torch/nn/functional.py", line 2267, in embedding
return torch.embedding(weight, input, padding_idx, scale_grad_by_freq, sparse)
RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper_CUDA__index_select)
```
On the other hand, in the case of invoke, the output is normal:
```
llm.invoke("Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World.")
```
```
'Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World. Hello World.\n\nThis is a simple program that prints the phrase "Hello World" to the console. \n\n**Here\'s how it works:**\n\n* **`print("Hello World")`**: This line of code uses the `print()` function, which is a built-in function in most programming languages (like Python). The `print()` function takes whatever you put inside its parentheses and displays it on the screen.\n* **`"Hello World"`**: The text within the double quotes (`"`) is called a string. It represents the message we want to print.\n\n\nLet me know if you\'d like to explore other programming concepts or see more examples! \n'
```
### Problem Analysis
- Apparently, I put kwargs in while generating pipelines and it applied
to `invoke()`, but it's not applied in the `stream()`.
- When using the stream, `inputs = self.pipeline.tokenizer (prompt,
return_tensors = "pt")` enters cpu.
- This can crash when the model is in gpu.
### Solution
Just use `self.pipeline` instead of `self.pipeline.model.generate`.
- **Original Code**
```python
stopping_criteria = StoppingCriteriaList([StopOnTokens()])
inputs = self.pipeline.tokenizer(prompt, return_tensors="pt")
streamer = TextIteratorStreamer(
self.pipeline.tokenizer,
timeout=60.0,
skip_prompt=skip_prompt,
skip_special_tokens=True,
)
generation_kwargs = dict(
inputs,
streamer=streamer,
stopping_criteria=stopping_criteria,
**pipeline_kwargs,
)
t1 = Thread(target=self.pipeline.model.generate, kwargs=generation_kwargs)
t1.start()
```
- **Updated Code**
```python
stopping_criteria = StoppingCriteriaList([StopOnTokens()])
streamer = TextIteratorStreamer(
self.pipeline.tokenizer,
timeout=60.0,
skip_prompt=skip_prompt,
skip_special_tokens=True,
)
generation_kwargs = dict(
text_inputs= prompt,
streamer=streamer,
stopping_criteria=stopping_criteria,
**pipeline_kwargs,
)
t1 = Thread(target=self.pipeline, kwargs=generation_kwargs)
t1.start()
```
By using the `pipeline` directly, the `kwargs` of the pipeline are
applied, and there is no need to consider the `device` of the `tensor`
made with the `tokenizer`.
> According to the change to use `pipeline`, it was modified to put
`text_inputs=prompts` directly into `generation_kwargs`.
## Issue
None
## Dependencies
None
## Twitter handle
None
---------
Co-authored-by: Vadym Barda <vadym@langchain.dev>
