cli/v1.2.0

BREAKING: Migrated job.resources.gpusPerTask to job.resources.gpus

gpus are now allocated for the whole job instead of "per task". This will allow heterogenous steps, example:

resources:
  tasks: 5
  gpus: 4

steps:
  - launch: # Launch async
      steps:
        command: echo "no gpu"

  - run: # 4 gpus
      resources:
        tasks: 4
        gpusPerTask: 1
      command: nvidia-smi

step.run.resources.gpusPerTask is still there and now defaults to 0 instead of inheriting the job.resources.gpusPerTask. This means that to migrate old jobs, this would require to explicitely add gpusPerTask to steps. Example:

Before:

resources:
  tasks: 4
  gpusPerTask: 1

steps:
  - run:
      command: nvidia-smi

After:

resources:
  tasks: 4
  gpus: 4

steps:
  - run:
      resources:
        gpusPerTask: 1
      command: nvidia-smi

SHA256 Checksums

8eee02b5c0ed0595a24a82ba1af3ee43041e228fb816d1a317fb6976ac78b884 *dps-darwin-amd64
7029c930ca7d50d71896f4cc055cd178f986b14f75c828dada6e500a0608ac62 *dps-darwin-arm64
66a3f578ee1e24b7a416acffb650d7cd65c564448dfa4299b322d1864edec1ee *dps-freebsd-arm64
66a3f578ee1e24b7a416acffb650d7cd65c564448dfa4299b322d1864edec1ee *dps-freebsd-arm64
a042d54ee0fe5205c7e9555301b68a2f8b744b4d1324a68ec513b4586b0c3dbb *dps-linux-amd64
69f975c66240ecb0058ba008eb3147bafc681226d1f90604d63a11a55e9b402b *dps-linux-arm64
b12c0a10632b8edfe23399f95ef72c79f705e169a72e2d1cdeb3544b05819f57 *dps-linux-riscv64
085d35d6898bc06eefc97a545b22db97ace5bfac955c92355a3c48da8b89c3b8 *dps-windows-amd64.exe