Feedback on 0.2 docs

[baskaryan]

We're releasing LangChain 0.2 the week of 5/20/24 (see #21437), and a major part of this release is revamping our docs.

The high level changes to the docs are that we:

• Are strictly following diataxis
• Have flattened the structure
• Support versioning

You can read a bit more about these changes in our 0.2 intro blog and see a preview of the 0.2 docs here.

We'd love your feedback on the 0.2 docs.

All of these changes have been informed by great feedback we've gotten on the existing docs, and we want to make sure that we're continuing to improve.

If you'd prefer to leave anonymous feedback, you can use this Google Form.

[nicpame]

Wonderful, I guess it would be better to have a more distinctive title for the tutorials menu in the more section like "3rd party tutorials" to distinguish it between (official) "tutorials" menu in the sidebar menu

[baskaryan]

Thanks for suggestion @nicpame, updated!

[PeerBoerner1]

I'd like to help with langchain documentation. What is the best way to determine where to focus my effort? Would it be to review the issues or just dig in and contribute where I feel there is the most need? I've not contributed to open source projects before so please forgive my ignorance.

[eyurtsev]

The best way to contribute to the main python docs (https://python.langchain.com/v0.2/docs/) in the immediate future is to provide feedback at the level of organization.

1. Is it organized better?
2. Are any important concepts missing?
3. Are there duplicates in documentation? Should we be removing something?

---

We always appreciate help with improving API reference docs with examples and improving in code documentation!

LangChain has relatively good coverage of Runnables, but a lot of other important code, schema and methods aren't documented with in code examples in the API Reference docs.

https://api.python.langchain.com/

For example, in code examples documented here:

https://api.python.langchain.com/en/latest/runnables/langchain_core.runnables.base.Runnable.html#langchain_core.runnables.base.Runnable

https://api.python.langchain.com/en/latest/runnables/langchain_core.runnables.base.Runnable.html#langchain_core.runnables.base.Runnable.pick

And are missing in many important objects:

https://api.python.langchain.com/en/latest/character/langchain_text_splitters.character.RecursiveCharacterTextSplitter.html#langchain_text_splitters.character.RecursiveCharacterTextSplitter

One process that could work for this:

1. Scan through tutorials, how-tos and concept guides here: https://python.langchain.com/v0.2/docs/ to identify important abstractions or implementations
2. Improve in code documentation for these.

[PeerBoerner1]

Many thanks for the guidance! I'll start reviewing as you suggest. I've always felt that we should provide more thorough documentation of the important objects, which is where I'm keenly interested to contribute.

[baskaryan]

Yep helping with API reference is super appreciated! #21983

[lqandmore]

When I exercise the demo with "Build a Chatbot->managing-conversation-history".
Messages history con't start with AIMessage.
the new messages if start with AIMessage after filter, will present an error:

Traceback (most recent call last):
  File "/Users/liquan/pythonDemo/LCDemo1/src/lcdemo1/chatbot.py", line 48, in <module>
    response = chain.invoke(
               ^^^^^^^^^^^^^
  File "/Users/liquan/pythonDemo/LCDemo1/.venv/lib/python3.12/site-packages/langchain_core/runnables/base.py", line 2368, in invoke
    input = step.invoke(
            ^^^^^^^^^^^^
  File "/Users/liquan/pythonDemo/LCDemo1/.venv/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 170, in invoke
    self.generate_prompt(
  File "/Users/liquan/pythonDemo/LCDemo1/.venv/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 599, in generate_prompt
    return self.generate(prompt_messages, stop=stop, callbacks=callbacks, **kwargs)
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
  File "/Users/liquan/pythonDemo/LCDemo1/.venv/lib/python3.12/site-packages/langchain_core/language_models/chat_models.py", line 456, in generate
    raise e
 .....
 ValueError: status_code: 400 
 code: InvalidParameter 
 message: Preprocessor error

the source code is:

def filter_messages(messages, k=10):
    #here maked the messages start with AIMessage
    return messages[-k:]

prompt = ChatPromptTemplate.from_messages(
    [
        (
            "system",
            "You are a helpful assistant. Answer all questions to the best of your ability in {language}.",
        ),
        MessagesPlaceholder(variable_name="messages"),
    ]
)

chain = (
    RunnablePassthrough.assign(messages=lambda x: filter_messages(x["messages"]))
    | prompt
    | chat
)

messages = [
    HumanMessage(content="hi! I'm bob"),
    AIMessage(content="hi!"),
    HumanMessage(content="I like vanilla ice cream"),
    AIMessage(content="nice"),
    HumanMessage(content="whats 2 + 2"),
    AIMessage(content="4"),
    HumanMessage(content="thanks"),
    AIMessage(content="no problem!"),
    HumanMessage(content="having fun?"),
    AIMessage(content="yes!"),
]
response = chain.invoke(
    {
        "messages": messages + [HumanMessage(content="what's my name?")],
        "language": "English",
    }
)

[efriis]

Hey there! I think different model providers handle a message history starting with AIMessage differently. Which model are you using, and if your model doesn't support that, I would recommend trimming your message history in a way that results in a HumanMessage first!

[lqandmore]

Thank you, I test with OpenAI model that's right, it's really the model previous make this problem.

[campio97]

Hi, I was trying to reproduce the code in the documentation related to agents. However, in the section (https://python.langchain.com/v0.2/docs/tutorials/agents/#tools), when selecting the Firework-AI example, the Mixtral-8x7b-instruct model is used, but it does not seem to be able to call the tools.

[abdurrahimcs50]

Hi LangChain Team,

I’ve had the pleasure of reviewing the 0.2 docs and I must say, I’m genuinely impressed! The implementation of versioning support is a significant advancement that greatly enhances the user experience. Additionally, the flattened structure has simplified navigation, making the documentation more user-friendly.

Overall, fantastic work on the docs!

Best,
MD Abdur Rahim

[Sameera2001Perera]

We're releasing LangChain 0.2 the week of 5/20/24 (see #21437), and a major part of this release is revamping our docs.

The high level changes to the docs are that we:

• Are strictly following diataxis
• Have flattened the structure
• Support versioning

You can read a bit more about these changes in our 0.2 intro blog and see a preview of the 0.2 docs here.

We'd love your feedback on the 0.2 docs.

All of these changes have been informed by great feedback we've gotten on the existing docs, and we want to make sure that we're continuing to improve.

If you'd prefer to leave anonymous feedback, you can use this Google Form.

@baskaryan I couldnt find any descriptive details about "ConversationSummaryBufferMemory" in the preview of the 0.2 docs. In 0.1 langchain memory classes are still in Beta. I have some issues with ConversationSummaryBufferMemory as I mentioned here. I just want to know if there is any fixes in 0.2.

[akshaylingamaneni]

i prefer the older structure, maybe due to my mental map of where things are. Are there plans of removing 0.1 docs soon?

[baskaryan]

No plans to remove 0.1 docs!

[efriis]

Haha I had a similar reaction on the mental map until I got used to the new structure!

If you think of features of the 0.1 docs that you think would improve the 0.2 docs, would love to hear them!

[yrobel-lima]

Same. It's not the mental map. This new structure is confusing, for example, in the How-to guide you get lost with so many examples on one page. In version 0.1, they were more accessible. The Langchain documentation is based on those examples.

[graphicmist]

I will suggest keep the overview structure similar for a longer time because as mentioned by @akshaylingamaneni it help build the mental map and familiarity on diff doc sections.

[DmitriyLeybel]

🙏Diataxis!
Looking good so far.

My favorite section:

In the words of Steve Balmer: DEVELOPERS DEVELOPERS DEVELOPER DEVELOPERS DEVELOPERS

[efriis]

Just updated a few things related to search indexing and ranking. If you have feedback about nonsensical search results, or queries that you think should map to specific pages, I would love to hear them to fix!

[eyurtsev]

Here's an example:

I think all of these queries should lead to the same page featuring the ChatOpenAI integration

openai chat
openai chat - note trailing space
openai chat model
openai chatmodel
chatmodel openai

[efriis]

Interesting - this one is probably because we rank "exact match" relatively high. Can experiment with demoting that a bit.

[akshaylingamaneni]

i have been thinking about this problem, people keep updating docs, how can we trace back? the feeling we have, i know its supposed to be here, where is it now?. llm's can tell with probability where it has been moved to based on search query similarity. addition's could be, if we have a dict of old hyperlink to new hyperlink can augment search as well. is this thought process makes sense?

would be interested knowing hows does docs migration works, the thought process behind it.

[svaeyens]

It would be great to be able download the full documentation as 1 pdf file. Makes it easier for me to create context for a custom bot with RAG.

[hinthornw]

For RAG, you could probably just downloda the ipynb notebook files from the repo to be fast

[jporter19]

Thumbs up on working on code examples, they're great when available. Suggestion: Langchain integrates with many other technologies, I think it would be useful to comment on the relationship between "langchain language" and the "integrated technology language". Using Langchain_chroma as an example: vectorstore = Chroma.from_documents(docs, OpenAIEmbeddings())
In chromadb the core language includes client(db) and collections. I think the code above makes an in memory collection, but what is the name of the collection? Can I append to an existing collection? Can I open an existing persistent client? Can I create a new client with a new collection? can I list all available collections in a persistent client (FYI..I think client.list_collections() is missing from your api)? Opening a client, choosing a collection, and add/retrieve documents to & from it seems foundational to all databases, but is not clearly documented. I see a lot of methods in langchain_chroma and much of what I'm asking may be available, but it's difficult to understand because I speak chroma and I'm learning langchain_chroma. Example: adding documentation stating "this line of code creates a chroma collection which we refer to as a vectorstore and can be named using the parameter collection_name= "name" and if you want the collection to persist use langchain commands....etc...