Platform
CRE™
Al-driven resource predictions
and recommendations
Resource Optimizations
SLA-based, lowest granular
dynamic resource allocations
Scheduler
Universal scheduler: Deploy compute clusters anywhere
Compute Optimization and Orchestration.
Launch your Private computecluster and run AI models anywhere
rapt AI CRE™
- Run-time resource predictions for your model workloads
- Infra Configurations
AI-driven Resource predictions and recommendations
- Get resource predicitions and recommendations – Compute type (Nvidia GPU, AMD GPU,CPU,TPU, etc.)
- Compute shares, Num of GPUs, Num of workers, Cost, Cloud, GPU mem, GPU cores, etc.
- Input your Model recipe
- Get ML based Resource Recommendations for your model workload
- API based – Add the rapt_predictions() API to your pipelines to receive resource recommendations.
Get Cloud Recommendations
Get best cloud, cloud instance type (spot, reserved, on-demand) and costs to run model based on model
AI-driven Model Configurations
Input model recipe, set activation recomputation method, sequence length and get right infra configuration results in one-click, No trial-and-error or configuration iterations.
- Lowest granular optimizations for AI models
- Allocates precise required GPU mem, SMSs, active blocks
- On-demand GPU shares based on workloads needs
rapt AI Resource Optimizations
Allocate Num of Cores, SMs and GPU mem dynamically based on model workload.
On-demand resource allocation dynamically in real-time while model runs. No disruptions.
rapt AI Scheduler
Launch AI Compute Cluster across any cloud and on-premise. Maximize your GPU clusters.
Auto Fractional GPU
Based on workload.
No Human Intervention.
No share presets required
Set SLAs (Cost, Perf).
Set Job Priorities (High, medium, value)
Use any cloud and on-premise.
Distribute your AI models across any Cloud. Cloud Flexibility.
Compute migrations.
Auto-manage spot disruptions and node failures.
Auto-preempt,
Suspend and resume AI models on GPUs
No Job Interference.
Auto-pack jobs without perf. impact to jobs.
