Key features in v1.8.0

Run step using GPU

A new model type cuda-python base on NVIDIA CUDA images has been added to allow a model execution on GPU instances. See the dedicated guide for more details.

Run metric rewok

Run quota is no more count in number of runs, but by using a new unit the credit.
A credit represent an infrastructure consumption (CPU, RAM or GPU) per minute:

• 1 credit —> 1 CPU / min
• 1 credit —> 4 Gi RAM / min
• 4 credit —> 1 GPU / min

Credit attibution

• A maximum credit is set following your namespace:
  • for DestinE free user maximum is 200.
  • for DestinE tenancy/project user maximum depends to the resource attached to your DestinE tenancy/project.
• Run credit are refill once per day
  • Maximum credit cannot be exceeded.
  • Extra exceeded credits from the previous day will be deducted during the refill.

Model pipRequirements parameter update

Model pipRequirements parameter update allows to provide scpecific options foreach requirement.

{
  "pipRequirements": [
    {
      "name": "torch",
      "version": "2.8.0+cu128",
      "options": [
        "--extra-index-url https://download.pytorch.org/whl/cu128"
      ]
    },
    {
        "name": "flash_attn",
        "version": "2.8.3",
        "options": ["--no-build-isolation"]
    }
  ]
}