v0.7.0

What's Changed

• Added batch-launch command to launch multiple models in a batch, currently this feature only supports single node (single/multiple GPUs within a single node) launches.
• Added cleanup command to remove logs using various filters
• Decoupled all the environment settings to an environment.yaml file, further simplifies migration process to different clusters
• Enabled RDMA/Infiniband support in the default container environment
• Adapted vec-inf package to Vector Killarney cluster:
  • Use apptainer instead of singularity
  • Set account and working directory as required fields, can also be set with environment variables
  • QoS, partition are now optional, added resource type for specifying type of compute
• Added env and config option for parsing additional environment variables and directly specifying a difference config to use for launch command
• Code base refactor, removed redundant code in generated scripts
• list command now sorts display by model name
• Added model tracking sheet for tracking cached model weights and model configurations on Killarney

New Contributors

• @scarere made their first contribution in #131

Contributors

@XkunW @scarere @kohankhaki