Serverless Training

Run code on GPUs without managing infrastructure.

Zero config. Per-second billing. Automatic hardware selection. Just submit your code.

lyceum-cli
$ lyceum python run train.py --machine gpu.h100
Uploading train.py...
Resolving dependencies...
[INFO] Job started on H100 (80GB)
Epoch 1/10: loss=2.341
Epoch 2/10: loss=1.892
[DONE] Job completed in 4m 32s. Cost: $0.26
.py
GPU

Zero configuration

No Dockerfiles, no provisioning, no cluster setup. Just submit your code and go.

2:34
$0.08

Per-second billing

Pay only for compute time used, not idle time. Billing stops the moment your job finishes.

T4
A100
H100
L40s

Automatic hardware selection

Lyceum picks the optimal GPU for your workload. Or specify exactly what you need.

Job done
Billing stopped

No idle costs

Machines spin down automatically when your job finishes. You never pay for unused compute.

GPU pricing

All GPUs billed per second. No minimum commitment.

GPU VRAM Price/hour Price/second
NVIDIA B200
192 GB $5.89 0.0982
NVIDIA H200
141 GB $3.69 0.0615
NVIDIA H100 Popular
80 GB $3.29 0.0548
NVIDIA A100
80 GB $1.99 0.0332
NVIDIA L40S
48 GB $1.49 0.0248
NVIDIA T4
16 GB $0.39 0.0065

Start training in under 60 seconds.

No infrastructure to manage. Just code and compute.