-
-
Notifications
You must be signed in to change notification settings - Fork 871
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Add: mlflow for experiment tracking #1059
Conversation
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Thanks for the PR!
@@ -694,6 +694,10 @@ wandb_name: # Set the name of your wandb run | |||
wandb_run_id: # Set the ID of your wandb run | |||
wandb_log_model: # "checkpoint" to log model to wandb Artifacts every `save_steps` or "end" to log only at the end of training | |||
|
|||
# mlflow configuration if you're using it | |||
# Make sure your `MLFLOW_TRACKING_URI` is set. | |||
mlflow_experiment_name: # Your experiment name |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
It might be worth logging a warning when this is set but the uri env isnt.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Wouldn't it be better to make a config mlflow_tracking_uri:
? The parsing code should be able to set that if needed.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I agree with what you propose @NanoCode012 . Have update the code.
Update mlflow_tracking_uri
Is there any need to pass this config to the HF Trainer to enable log? I think some validation test would also be needed to make sure both aren't active at same time. In addition, would you be able to provide an example run output + image on mlflow? |
I disagree, isn't the tracking uri equivalent to a secret token? It can get inadvertently exposed if you share your YAML (say integration with wandb also) |
To my understanding it is only an url actually. In Mlflow auth seams to be handled with username and password. But I might miss something. For reference https://mlflow.org/docs/latest/auth/index.html I have no strong opinion if it should be or not be in the config file. |
Oh, I wasn't aware it was a secret. In this case, I would not recommend it being in the yaml anymore due to reasons wing mentioned. |
update trainer building
looks like the URI isn't considered a secret, but could be sensitive if someone starts a server and doesn't enable auth on it. |
Adding MLFOW to Axolotl for experiment tracking, looked into how Weight and Bias has been setup and tried to follow the same pattern. Have tested the changes and everything looks good to me.
Happy for any feedback or comments.