dstack

dstack copied to clipboard

[Bug]: No graceful shutdown when building Packer images

5

### Steps to reproduce 1. Run the "Build docker & cloud images" GitHub Action. Uncheck "Build docker images", check "Staging build" and one of the clouds. 2. Wait until the...

jvstme

[UX] Support a shorter syntax for model mapping

2

Currently, I have to specify: ``` model: type: chat name: meta-llama/Meta-Llama-3.1-70B-Instruct format: openai ``` I suggest that we allow to specify ``` model: meta-llama/Meta-Llama-3.1-70B-Instruct ``` In that case `type` and...

peterschmidt85

[Feature]: Generate Docker registry password using AWS creds

4

### Problem In order to use private Docker images hosted with AWS ECR, one has to [generate](https://docs.aws.amazon.com/AmazonECR/latest/userguide/registry_auth.html) a temporarily password using `aws ecr get-login-password` and then hardcode them in the...

peterschmidt85

[Feature]: Provide an option to delete old run logs automatically

3

### Problem Run logs are never deleted, so disk space usage on the server can grow quickly and indefinitely. ### Solution Add a setting that will specify the TTL for...

jvstme

[Feature]: Better logging when a run terminates due to max_duration

2

### Problem In the server console logs, it's unclear that a run was terminated due to max_duration. Attaching image showing when a run was started and the messaging shown 6...

james-boydell

[Feature]: Enable all available network cards on AWS instances

2

### Problem Currently, with the AWS backend `dstack` unconditionally requests one network interface, even with instance types that have multiple network cards (e.g., `p5.48xlarge` has 32 EFA-capable cards). Network performance...

un-def

[Bug]: DinD doesn't allow passing `--gpus` flag

2

### Steps to reproduce 1. make a `repro.dstack.yml` with: ```yaml type: task name: my-repro-task image: dstackai/dind:latest privileged: true commands: - start-dockerd - sleep infinity resources: cpu: 4.. memory: 6GB.. gpu:...

mtaran

Volumes can be supported via [File Systems](https://lambdalabs.com/blog/persistent-storage-beta). File Systems are persistent disks that can be mounted at fixed mount points of one or more VMs. It’s not possible to use...

r4victor

[Bug]: Adding on-prem node/instance to two different fleets shouldn't be allowed

1

### Steps to reproduce Create a fleet with an on-prem node/instance. Change the name of the fleet in the config and apply. The on-prem node will now be a part...

james-boydell

peterschmidt85