mlflow.langchain.log_model : AttributeError: module 'langchain' has no attribute '__version__'

[Abonia1]

Having issue while using mlflow.langchain.log_model in databricks:
Langchain version = 0.0.125

mlflow.langchain.log_model(
    lc_model=llm_chain,
    artifact_path="model",
    registered_model_name="flan-t5"
)

WARNING mlflow: MLflow does not guarantee support for LLMChains outside of HuggingFaceHub and OpenAI, found <class 'langchain.llms.huggingface_pipeline.HuggingFacePipeline'>
AttributeError: module 'langchain' has no attribute 'version'

thanks in advance.

[WeichenXu123]

Hi @Abonia1 Could you provide your langchain package version and the complete reproducing code ? Thank you!

[Abonia1]

Hello @WeichenXu123 ,thanks for taking a lead on this topic.
As I got the error with latest version of langchain I have repoted the issue here and using langchain==0.0.125.

Complete code :

from langchain import PromptTemplate, HuggingFaceHub, LLMChain
from langchain.llms import HuggingFacePipeline
from transformers import AutoTokenizer, AutoModelForCausalLM, pipeline, AutoModelForSeq2SeqLM


import torch
import mlflow

tokenizer = AutoTokenizer.from_pretrained("google/flan-t5-small")

model = AutoModelForSeq2SeqLM.from_pretrained("google/flan-t5-small")
pipe = pipeline(
    "text2text-generation",
    model=model, 
    tokenizer=tokenizer, 
    max_length=100
)

local_llm = HuggingFacePipeline(pipeline=pipe)
template = """Translate everything you see after this into French:{input}"""
prompt = PromptTemplate(template=template, input_variables=["input"])

llm_chain = LLMChain(prompt=prompt, 
                     llm=local_llm
                     )
question = "What is the capital of England?"
print(llm_chain.run(question))

mlflow.langchain.log_model(
    lc_model=llm_chain,
    artifact_path="model",
    registered_model_name="flan-t5"
)

[WeichenXu123]

I thin you are using an out-of-date langchain package , the latest version is 0.0.170 now. And I tested in 0.0.170 version we don't have the issue.

[Abonia1]

But with recent version I can't import Langchain in databricks. I have reported the issue and its still open here .

As old version is supported by databricks now, there is any work around to make mlflow work with langchain==0.0.125 ?

Any help would be appreciated.

[WeichenXu123]

Got it. Let me check. I think we need to address the langchain issue on databricks

[WeichenXu123]

@Abonia1 Could you try this ? langchain-ai/langchain#4401 (comment)

[mlflow-automation]

@BenWilson2 @dbczumar @harupy @WeichenXu123 Please assign a maintainer and start triaging this issue.

[Abonia1]

@WeichenXu123 Even I have tried

%pip install langchain typing_extensions==4.1.0

I got this error as we don'thave access to model in community version of databricks:

For info: As mentionned here, there are certain limitations with the community edition and we do not have this feature there. To use this we need to go with the commercial version of Databricks.

So consider we can't use the feature of mlfow+langchain in community version of databricks. Thanks for all the help.

[WeichenXu123]

@Abonia1

Sorry my previous comment is wrong, I updated my comment, langchain-ai/langchain#4401 (comment)
you need to install typing-extensions>=4.2.0 to address the issue.

For info: As mentionned here, there are certain limitations with the community edition and we do not have this feature there. To use this we need to go with the commercial version of Databricks.

Yes this is an limitation