Developer Guide Community Github

Create Task

All users could create collaborative training tasks in two ways: testnet.flock.io or FLock client.

Detailed guidance on the following:

  1. Define your task details:

  • Name: define a specification task name that you want to do. The name should be simple and straightforward. For example: Finance Analyst LLM Finetuning

  • Description: try to describe more details about this task, for example, This task aims to fine-tune a Finance Analyst LLM based on Vicuna V1.5, which can provide professional ideas in the finance field. In this task, we expect all participants to provide proper and high-quality finance data, including, but not limited to, finance news, trading records, and real-time stock data.

  • Input: describe what type of data will be used for training, for example, Text

  • Output: describe what type of data will be responded to from the model. For example, Text

  • Model Name: select a model that will be selected for this training task, for example, Vicuna V1.5

  • Task Type: select a type that can present this task. For example, LLM Finetuning

  • Model Size in Memory: roughly provide how big will be the model saving in memory. For example, 13GB

  • Output Description: describe what kinds of information that user can get from the model. For example response of LLM

  • Model Definition Hash: this is the IPFS address of the task container image. Task creators can get it by themselves (refer to https://github.com/FLock-io/v1-sdk) or directly use an existing one. The structure looks like Qma3Wba4pJvmduyarsLznEwXVYzyqysy5YZJjk4e9M4nNL

  1. Define the training dataset

  • Prepare and upload a small size of sample data, our system will detect the data schema automatically. Alternatively, user can change the schema manually.

  1. Define the training settings

  • Limit minimum number of participants: select how many participants you need to start this task. For example 2

  • Limit maximum number of participants: select how many participants you want to limit in this task totally. For example 4

  • Number of training rounds in total: define how many rounds you want for this training task. For example 200

  • Initial stake requirement: define how much $FLC user needs to stake to participate in this task. For example 20

  • Rewards pool: define how much $FLC in total you want to reward to all participants. For example 2000

  1. Check the created task in the task list in the dashboard of testnet.flock.io or FLock client

  1. Get your task unique address on the detail page of a task. For example, Task address: 0xD680C654C3Bcb1FcA351254ac033681D98650bC4

PreviousLocal TrainingNextHow to participate

Last updated