Skip to main content

Key Concepts

Key Concepts: Job

Overview:

In the realm of LLMP, a "Job" signifies a distinct task or operation to be performed. It's an encapsulation of a set of instructions, examples, and associated data, managed through the system to produce desired outputs. A job is the primary entity that users interact with, effectively communicating their requirements to the LLMP system.

What makes up a Job:

  1. Definition:

    • Before a job starts, it needs a clear definition. This includes what the job aims to achieve, its input and expected output structure, and any specific instructions that the LLM should follow.
  2. Lifecycle:

    • A job has a lifecycle that begins with its creation and ends with its completion. During this lifecycle, the job can undergo modifications, can be paused, or even terminated based on the user's needs.
  3. Versioning:

    • As the system or user's requirements evolve, so do the jobs. To accommodate this, jobs in LLMP support versioning. This means any significant changes to a job can be tracked with version numbers, allowing users to rollback or reference specific versions if needed.
  4. Logs & Metrics:

    • Keeping track of a job's progress, successes, and failures is crucial. LLMP maintains logs associated with each job and may also generate metrics to provide insights into the job's performance and outcomes.
  5. Templates & Records:

    • Jobs in LLMP are structured using templates that standardize their format and make them reusable. As jobs are executed, records of their activity, such as the JobRecord, are maintained, offering detailed insights into each job's attributes and history.

Why "Job" is Central to LLMP:

Jobs are the interface between the user and the LLMP system. They take abstract requirements and convert them into actionable tasks within the system. Understanding the concept of a job and its intricacies is pivotal for anyone aiming to make the most of LLMP.

In the following sections, we will dive deeper into specific components related to Jobs, such as the JobRecord, to further explore their structures and functionalities.