Langchain issue

[nmnikhil]

File \Python\Python313\site-packages\nemoguardrails\actions\llm\utils.py", line 19, in
from langchain.base_language import BaseLanguageModel
ModuleNotFoundError: No module named 'langchain.base_language'
why am i getting this error is there any way to resolve this ?

[kamilmuratyilmaz]

Hello @nmnikhil,

I had the same error before, you should import BaseLLM, docs might be outdated.

Reference Code

[trebedea]

Hi @nmnikhil ,
Can you share some details about when you get that error?

[trebedea]

Sure, it would be great to help.
Langchain moved the package for BaseLanguageModel in more recent versions to langchain_core.language_models.base.BaseLanguageModel (at least in 0.3.43). Doesn't it work if you import it from the correct package?

[kamilmuratyilmaz]

When we installed guardrails with the develop branch ( which supports Python 3.12) utilizing poetry, package manager installed Langchain 0.3.20. Maybe this is the main reason behind the issue, our team haven't tested this out.

[trebedea]

Looking more into this, it's strange the error you get because for me the following works with langchain-0.3.20 langchain-core-0.3.43 :

from langchain.base_language import BaseLanguageModel

Can you pip freeze the langchain version?

[kamilmuratyilmaz]

@trebedea It is really strange indeed, tried with pip freeze again, it is 0.3.20.

We are having this error;

TypeError: Can't instantiate abstract class BedrockClient without an implementation for abstract methods 'agenerate_prompt', 'apredict', 'apredict_messages', 'generate_prompt', 'invoke', 'predict', 'predict_messages'

Yet after we utilized _generate method shown like here and importing BaseLLM instead of BaseLanguageModel it works.

[trebedea]

@kamilmuratyilmaz , you are right. Actually it seems langchain_core.language_models.llms.LLM is the most similar in spirit to the old BaseLanguageModel. LLM extends BaseLLM, details here: https://python.langchain.com/api_reference/core/language_models/langchain_core.language_models.llms.LLM.html

This is also how Langchain suggests in their documentation to create a custom LLM: https://python.langchain.com/docs/how_to/custom_llm/
But I'm also fine with extending BaseLLM in the docs if you think this is better. We also extended BaseLLM for our custom LLMs (e.g. for NemoLLM), but I do not see right now any advantage for not using Langchain's recommendation.

If you can help with a MR on changing the documentation, this would be great.