Compute network
work to do.
GPAIR's workload system is designed around provider-confirmed execution and measured usage. Job submission remains unavailable until a real runner is connected and tested.
Execution not enabled
No workload button is exposed while runtime support is unavailable. Existing workload records appear on their coin's Compute Profile.
Runtime capability
Live system status
Job state machine
A measured path from queue to result
Queued
Accepted and waiting for capacity
Starting
Provider resource and runtime starting
Running
Provider confirms active execution
Completed
Finished with measured usage
Failed
Stopped with a recorded failure
Cancelled
Ended by an authorized owner
Workload categories
Built into the product model
AI inference
Serve prompts to an LLM or vision model
Runtime not enabled
Image generation
Run diffusion models on demand
Runtime not enabled
Video generation
Render short AI video clips
Runtime not enabled
Model hosting
Keep a model online behind an endpoint
Runtime not enabled
Training
Train models from scratch
Runtime not enabled
Fine-tuning
Adapt an open model to your data
Runtime not enabled
Community jobs
Holders or users submit compute jobs
Runtime not enabled
Start with the pair