Thread by @zongheng_yang

Introducing SkyPilot:

Run ML and Data Science jobs on any cloud, with massive cost savings.

🚀 Run jobs on any cloud
⏰ Get GPU/TPU/CPU in 1 click
💵 Reduce > 3x cost

skypilot.co
Read blog: medium.com/@zongheng_yang…

🧵1/

2/ Clouds are expensive 💵. SkyPilot reduces cloud bills by 3-6x.

Here's SkyPilot training BERT on WikiText-103 from @huggingface on spot instances *across* AWS & GCP. (Video attached above)

SkyPilot offers auto-recovery from preemptions, saving 3x cost for this job.

3/ Save more 💵💵💵 by (1) auto stopping idle clusters (2) auto picking cheapest region/cloud.

Here's an example of SkyPilot shopping for the cheapest region/cloud for allocating 8x A100 GPUs.

Single-cloud users also save on intra-cloud price differences (see blog for details).

4/ Clouds are also getting hard to use 🧐. With SkyPilot, you can run ML/Data Science jobs on any cloud, with no code change 🚀.

SkyPilot simplifies the cloud infra heavy-lifting that ML/DS users don't want to deal with.

Running an existing project is simple:

5/ Reliably spin up GPU/TPU/CPU

Clouds often run out of capacity and getting VMs fails.

SkyPilot solves this by:
🌐 checking all regions and clouds
🔀 auto-failover across them

⏰ Reliably spin up a VM with 1 command:
$ sky gpunode
$ sky tpunode
$ sky cpunode

SkyPilot is in active use by dozens of ML/data science users in 10+ orgs.

• Groups at @berkeley_ai @stanfordnlp use SkyPilot for training on GPUs and TPUs
• @salkinstitute uses SkyPilot to cut 80% cost for bio batch jobs on CPU instances
• Active collab w/ several companies

🚀🚀 Start using the cloud easily and cost effectively today 🚀🚀

$ pip install skypilot

Star our GitHub: skypilot.co
Read blog: medium.com/@zongheng_yang…

Follow @skypilot_org for new updates we'll be sharing in the coming months!

#Cloud #MultiCloud #ML #DataScience