Run GPU training jobs on OneInfer Cloud
Bring a script or a config. Pick the GPUs. Submit.
Training on OneInfer Cloud runs your fine-tuning and training work as batch jobs on the GPUs you choose. Upload a Python or shell script, or fill in guided fields: model, dataset, hyperparameters and optional LoRA or QLoRA settings.
Pick the GPU, GPU count, disk size and spot preference, check the estimated cost, and submit. The Console shows each job’s status and resources, and you can cancel queued or running jobs.
A training job in 5 steps
In the Console, open Training › Train / Fine-tune, choose New run › Run your own script and give the job a name.
The controls needed to submit and manage training work
The training page is connected to the same GPU, storage, and job-management workflows already available in OneInfer.
Use a Python training script
Upload your own Python script when you need complete control over the training code and runtime arguments.
Configure model and dataset inputs
Use the guided form to provide model and dataset paths, training parameters, and optional LoRA settings.
Choose the compute for each job
Select the GPU model and quantity, disk size, spot preference, and preferred cloud capacity for the workload.
Track the job lifecycle
See submitted jobs, refresh their latest status, review assigned resources, and cancel queued or running work.
Keep data and artifacts available
Use OneInfer storage for persistent datasets, checkpoints, and model artifacts that need to survive beyond a job.
Request clustered capacity
For workloads that need coordinated GPU capacity, submit a cluster request from the GPU marketplace workflow.
From configuration to a running batch job
Each job keeps its training configuration and compute selection together, so the console can show what was requested and what is running.
Choose how to configure the job
Start with an uploaded Python script or use the guided parameter form.
Add training inputs
Provide the model, dataset, hyperparameters, and optional LoRA configuration.
Select GPU resources
Choose the GPU type, count, storage, spot preference, and cloud preference.
Submit and monitor
Follow the job status from the console and cancel queued or running work when needed.
Keep training data and deployment work in one platform
Use persistent storage for datasets and model artifacts. When a supported custom model is ready, continue through the model deployment workflow to prepare it for inference.
Submit your first GPU training job
Open the console to configure resources, submit a batch job, and monitor its status.