1921 & GSK-1851 & GSK-1901 - save langchain chain without ml flow to reduce compatibility

[kevinmessiaen]

Description

Save and load LangchainModel using langchain provided methods instead of ml_flow

Related Issue

Type of Change

• 📚 Examples / docs / tutorials / dependencies update
• 🔧 Bug fix (non-breaking change which fixes an issue)
• 🥂 Improvement (non-breaking change which improves an existing feature)
• 🚀 New feature (non-breaking change which adds functionality)
• 💥 Breaking change (fix or feature that would cause existing functionality to change)
• 🔐 Security fix

[linear]

GSK-1921 Save langchain `Chain` without ml_flow to reduce compatibility issues

[linear]

GSK-1851 Add support for QA retriever upload in LLM Hub

GSK-1901 solve serialisability issue

from mattbit:

import giskard as gsk
import pandas as pd
from pathlib import Path


def model_fn(df):
    return [True] * len(df)


class MyCustomModel(gsk.Model):
    def save_model(self, path):
        print("SAVING CUSTOM MODEL TO", path)
        Path(path).joinpath("custom_data").touch()

    classmethod
    def load_model(self, path):
        print("LOADING CUSTOM MODEL FROM", path)
        model = lambda x: [True] * len(x)
        return model

    def model_predict(self, df: pd.DataFrame):
        return self.model(df)


print("TESTING MODEL SAVE")
gsk_model = MyCustomModel(model_fn, model_type="regression")
print("TYPE:", type(gsk_model))
Path("test_save").mkdir(exist_ok=True)

gsk_model.save("test_save")

print(MyCustomModel.load("test_save"))

[rabah-khalek]

the idea of factoring out save_model from save is that you shouldn't have to overwrite save, save_model is by default always called by BaseModel.save so I think you're calling it twice in this.

[kevinmessiaen]

Well actually save_model isn't called in neither WrapperModel nor BaseModel

[rabah-khalek]

Oh, you're right, my bad. I relied on my memory instead of looking at the code. I thought we had the same logic as model_predict and predict, but indeed we kept save loose I think because of the ordering of the saving.

[rabah-khalek]

These backward incompatible API changes from LangChain make me nervous. Could you please make sure that our code works for all the versions of LangChain we support in pyproject.toml?

[kevinmessiaen]

It's actually a bug fix. The former method only works with LLMChain while the new one works with any type extending Chain, this will continue to work with models uploaded with previous version

[rabah-khalek]

Is it related to this as well you think? https://linear.app/giskard/issue/GSK-1819/set-a-minimum-on-langchains-version

[rabah-khalek]

Looks good to me now @kevinmessiaen, thanks!

[sonarqubecloud]

Kudos, SonarCloud Quality Gate passed!   

0 Bugs
0 Vulnerabilities
0 Security Hotspots
0 Code Smells

0.0% Coverage
0.0% Duplication