SkyPilot is a framework for running AI and batch workloads on any infra, offering unified execution, high cost savings, and high GPU availability.
On Jan 27, 2025, DeepSeek AI released the Janus. It outperforms state-of-the-art Vision Language Models such as LLaVA, supporting a variety of Vision-Language tasks such as image generation and Q&A.
This guide walks through how to run and host models on any infrastructure from ranging from Local GPU workstation, Kubernetes cluster and public clouds (15+ clouds supported).
Install SkyPilot on your local machine:
pip install 'skypilot-nightly[all]'Pick one of the following depending on what infra you want to run:
If your local machine/cluster has GPU: you can run SkyPilot directly on existing machines.
If you want to use Clouds (15+ clouds are supported): See docs for details.
Now it's time to run Janus with SkyPilot. Commands may vary based on the GPUs available to you.
Run Janus (1.5B) with:
sky launch janus_1.5b.yaml \
-c janus \
--env HF_TOKEN=YOUR_HUGGING_FACE_API_TOKEN
or run Janus Pro (7B) with:
sky launch januspro_7b.yaml \
-c janus \
--env HF_TOKEN=YOUR_HUGGING_FACE_API_TOKEN
It will show all the available machines that you have and pricing
...
Considered resources (1 node):
-----------------------------------------------------------------------------------------------------------------
CLOUD INSTANCE vCPUs Mem(GB) ACCELERATORS REGION/ZONE COST ($) CHOSEN
-----------------------------------------------------------------------------------------------------------------
Kubernetes 4CPU--16GB--1L4 4 16 L4:1 gke-cluster 0.00 ✔
RunPod 1x_L4_SECURE 4 24 L4:1 CA 0.44
GCP g2-standard-4 4 16 L4:1 us-east4-a 0.70
AWS g6.xlarge 4 16 L4:1 us-east-1 0.80
AWS g5.xlarge 4 16 A10G:1 us-east-1 1.01
Fluidstack A100_PCIE_80GB::1 28 120 A100-80GB:1 ARIZONA_USA 1.80
RunPod 1x_A100-80GB_SECURE 8 80 A100-80GB:1 CA 1.99
Paperspace A100-80G 12 80 A100-80GB:1 East Coast (NY2) 3.18
Azure Standard_NV36ads_A10_v5 36 440 A10:1 eastus 3.20
Azure Standard_NC24ads_A100_v4 24 220 A100-80GB:1 eastus 3.67
GCP a2-ultragpu-1g 12 170 A100-80GB:1 us-central1-a 5.03
Azure Standard_ND96asr_v4 96 900 A100:8 eastus 27.20
GCP a2-highgpu-8g 96 680 A100:8 us-central1-a 29.39
AWS p4d.24xlarge 96 1152 A100:8 us-east-1 32.77
-----------------------------------------------------------------------------------------------------------------
You should be able to access directly via a terminal prompt via
Running on public URL: https://xxxxxx.gradio.live
or you can access through getting the IP address of the deployed instance via
echo `sky status --ip janus`
A blue sky, vast fluffy clouds surrounding, enchanting, immortal, dynamic motion, cinematic, Unreal Engine 5 and Octane Render, highly detailed, photorealistic, natural colors, epic atmosphere, and breathtaking realism
