[BFCL] Add the option to log to WandB during bfcl evaluate

[ThomasRochefortB]

In our internal use of BFCL at Valence Labs, we use WandB extensively to centralize the results of our benchmarking.
This PR adds a --wandb-project CLI argument to the bfcl evaluate command to upload the generated .csv files to wandb. I wanted to open this PR to see if this would be an feature of interest for BFCL :)

Here is an example command:

bfcl evaluate --model gpt-3.5-turbo-0125  
--wandb-project <wandb_entity:wandb_project>

This will log the data_live.csv, data_non_live.csv and data_overall.csv as dataframe artifacts on the WandB wandb_project under the wandb_entity.

[HuanzhiMao]

Thanks for the PR @ThomasRochefortB !

Two suggestions:

1. We should add wandb to the pyproject.toml; otherwise the pipeline will break. Maybe as an optional-dependency since wandb is not required for evaluation to complete?
2. Maybe we should add the wandb_project to an environment variable (in the .env file) and just use a CLI flag to signal that the result should be added to the wandb because it's likely user would use the same wandb_project across runs. So they don't need to type it in the command line everytime, and it makes the code cleaner.

What do you think?

[ThomasRochefortB]

Thank you for the feedback @HuanzhiMao !

I think these are 2 great suggestions. I will adjust my branch accordingly.

[HuanzhiMao]

Thank you for the feedback @HuanzhiMao !

I think these are 2 great suggestions. I will adjust my branch accordingly.

Sounds good!

[ThomasRochefortB]

There you go @HuanzhiMao ! I have made the following changes:

• Added an optional env var WANDB_BFCL_PROJECT in the .env.example
• Modified the init_wandb function to read from the env var
• Added Wandb as an optional dependencies on the pyoroject.toml

[HuanzhiMao]

LGTM
Thanks for the PR @ThomasRochefortB !