Seung Jin

[config] remove omegaconf as dependency

2

The only logic that needs to be implemented to remove omegaConf is the dotlist parsing logic. This actually turned out to be pretty simple to implement using `yaml.safe_load()`. Added some...

[k8s] idea: allow an accelerator to map to multiple label values

2

I was looking at `GFDLabelFormatter` and found this comment snippet ``` This LabelFormatter can't be used in autoscaling clusters since accelerators may map to multiple label, so we're not implementing...

[AWS] ability to specify transient per-cluster security group

4

Related issue: https://github.com/skypilot-org/skypilot/issues/3688 Mini-proposal: The issue raises a good point about how we deal with security groups in AWS. Currently, if the user doesn't specifically designate a security group to...

[k8s][gcp] Accept non-k8s TPU names

2

Running `sky launch` with TPUs on GCP support the following TPU names: ``` GOOGLE_TPU AVAILABLE_QUANTITIES tpu-v2-8 1 tpu-v3-8 1 tpu-v4-8 1 tpu-v4-16 1 tpu-v4-32 1 tpu-v5litepod-1 1 tpu-v5litepod-4 1 tpu-v5litepod-8...

[k8s] CPU only jobs to prefer nodes without GPUs

1

Problem statement: In a k8s cluster with multiple node pools, it is possible to have a heterogenous cluster where some nodes are CPU instances and some nodes are GPU instances....

[storage] storage upload failing due to "Argument list too long"

13

 `/bin/bash: /opt/homebrew/bin/aws: Argument list too long` Investigate the following error message when uploading files to cloudflare R2 storage

source : https://github.com/skypilot-org/skypilot/pull/5303#discussion_r2057023025 While working on https://github.com/skypilot-org/skypilot/pull/5303#discussion_r2057023025, it's become apparent that we don't really document what conda environment is set up where (and why), and it is quite confusing to...

python executable not found when running a docker image with Runpod

1

``` service: readiness_probe: /healthcheck replica_policy: min_replicas: 2 max_replicas: 12 target_qps_per_replica: 10 resources: image_id: docker:pytorch/pytorch:2.6.0-cuda12.4-cudnn9-runtime cloud: runpod region: US ports: 8000 cpus: 4+ memory: 8+ disk_size: 32 accelerators: - RTX4090 use_spot:...

Tested (run the relevant ones): - [ ] Code formatting: install pre-commit (auto-check on commit) or `bash format.sh` - [ ] Any manual or new tests for this PR (please...

Mini-proposal: As part of the effort to declutter the `~/.sky` directory, we should change the `privacy_policy` to a dotfile. This file really serves as an internal flag with no real...