Schedule
Last updated
Last updated
The Scheduled Task feature allows users to configure tasks to be executed at specific times, enabling efficient use of resources and minimizing disruption during active working hours. Here are the key aspects of this feature:
To add a new scheduled task:
Click on the icon located in the top-left corner of the interface.
Select the specific task you want to schedule.
Configure the corresponding training parameters, such as the model, dataset, batch size, and other details.
Save the settings to finalize the schedule.
Another way to add a scheduled task is through the Finetune History of a specific task. Users can select previous training parameters from the history and create a schedule based on those settings.
Tasks can be set up with various parameters including:
Model: Specify the model used for the task (e.g., google/gemma-2-9b-it).
Dataset: Input dataset for training or evaluation (e.g., AIR-Product.json).
Batch Size: Define the batch size (e.g., 8).
Total Batch Size: Define the batch size (e.g., 128).
Maximum Sequence Length: Configure the sequence length (e.g., 512).
Learning Rate: Set the learning rate (e.g., 0.00007).
Epochs: Specify the number of training epochs (e.g., 2).
Users can schedule tasks to start at specific times, such as:
During non-peak hours (e.g., evening after work or weekends).
For tasks requiring extended durations, scheduling prevents overlap with active operations.
Each scheduled task provides a summary of essential information:
Start Time: Indicates the exact time the task will begin.
Created/Updated Timestamp: Logs when the task was created or last updated.
Action Options: Provides actions such as editing or deleting the scheduled task.
Scheduling ensures that tasks like model training, which often require significant computational resources, are executed during idle periods. This minimizes system load during peak hours and maximizes resource efficiency.
