Deployment error: module 'mlflow.langchain' has no attribute 'output_parsers'

[daniel-hain]

Hello,

I'm running into a model deployment error in the following notebook:
genai_cookbook/10-min-demo/Mosaic-AI-Agents-10-Minute-Demo.ipynb
Everything works fine until the following line:
with mlflow.start_run(run_name="databricks-docs-bot"): logged_chain_info = mlflow.langchain.log_model( lc_model=os.path.join( os.getcwd(), f"{QUICK_START_REPO_SAVE_FOLDER}/quick_start_demo/sample_rag_chain", ), # Chain code file from the quick start repo model_config=chain_config, # Chain configuration set above artifact_path="chain", # Required by MLflow input_example=input_example, # Save the chain's input schema. MLflow will execute the chain before logging and capturing its output schema. )

I always get the error message:
MlflowException: Failed to run user code from /tmp/tmp_2mpqvqb/model.py. Error: module 'mlflow.langchain' has no attribute 'output_parsers'

Setup:

• Everything run within my databricks workspace
• MLflow version '2.15.1'

Anyone can help with that? Thanks a lot!

[fgraichen]

I ran into this as well. What version of the mlflow libraries are you using. When I upgraded to 2.15.1 (latest) this message went away, unfortunately I received a different error - log_model() got an unexpected keyword argument 'model_config' after upgrading the library. Curious if you experience the same

[daniel-hain]

Hi! I tried the same, and updated to MLflow version '2.15.1', but get the same error message.

[bbqiu]

hey all, please try pulling from latest master again -- we will be fixing this and some other small bugs with ChatCompletionOutputParser w/ the next release of mlflow.

specifically, make sure you have 8e120e4