Skip to content

Conversation

mayoor
Copy link
Member

@mayoor mayoor commented Apr 24, 2024

In certain datasets, the packing of dataset can cause failure during Fine Tuning. By allowing for turning off this option, we can unblock the cases where this occurs. We will enable only CLI for now. We can add more options on UI under advanced section.
We have to move batch_size to AquaFineTuningParms class.
We also need a method on the dataclass that can automatically turn the parameters to cmd arguments and avoid hard-coding.

Usage -

ads aqua fine_tuning create \
		--ft_source_id $(MODEL_SOURCE) \
		--ft_name $(NAME)-FT \
		--dataset_path  $(DATASET)\
		--report_path $(OUTPUT) \
		--ft_parameters '{"epochs": $(EPOCH), "learning_rate": 0.0002, "sample_packing": "$(SMP_PKG)"}' \
		--shape_name $(SHAPE) \
		--replica $(REPLICA) \
		--validation_set_size 0.1 \
		--subnet_id $(SUBNET)\
		--log_group_id $(LOGGRP) \
		--log_id $(LOG) \
		--experiment_id $(MVS) \
		--compartment_id $(COMPARTMENT)```

@oracle-contributor-agreement oracle-contributor-agreement bot added the OCA Verified All contributors have signed the Oracle Contributor Agreement. label Apr 24, 2024
@mayoor mayoor changed the base branch from main to feature/aquav1.0.1 April 24, 2024 16:34
@mayoor mayoor merged commit 2f3c48a into feature/aquav1.0.1 Apr 24, 2024
Copy link

📌 Cov diff with main:

Coverage-1%

📌 Overall coverage:

Coverage-21.35%

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
OCA Verified All contributors have signed the Oracle Contributor Agreement.
Projects
None yet
Development

Successfully merging this pull request may close these issues.

4 participants