skills.rest
hao-ai-lab
hao-ai-labOFFICIAL
Explore all skills by this author →
Website𝕏@haoailabGitHub

dreamverse-deploy

Redeploys the Dreamverse backend and frontend on a chosen local GPU with readiness checks.

4.3k|431|Updated Oct 24, 2024
Search Tags:#next.js#FastAPI#GPU#FFmpeg#Deployment#NVENC#devOps
One-click install
npx skills add https://github.com/hao-ai-lab/FastVideo --skill dreamverse-deploy
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: dreamverse-deploy
Source: https://github.com/hao-ai-lab/FastVideo/tree/main/.agents/skills/dreamverse-deploy
Command: npx skills add https://github.com/hao-ai-lab/FastVideo --skill dreamverse-deploy

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill requires flashinfer-python, dreamverse-server, and includes scripts (resource) components.

What problem does it solve?

Redeploying the Dreamverse app on a dev node involves killing stale processes on occupied ports, freeing GPU memory from orphaned workers, exporting the correct environment recipe, and waiting for backend and frontend readiness — a manual sequence that is error-prone and often leads to GPU OOMs from leftover processes.

Core Features & Use Cases

  • Automated teardown and relaunch: Kills processes on the target backend/frontend ports and any process holding the target GPU, including orphaned multiproc_executor workers, before booting fresh services.
  • Configurable startup modes: Optional GPU warmup, max-autotune torch.compile, and NVENC hardware encoding via flags or environment variables, with readiness budgets scaled to the chosen options.
  • Readiness verification: Polls the backend /readyz endpoint and the frontend root until both return 200, then prints URLs, PIDs, and log paths.
  • Use Case: After pulling new code for the migrated apps/dreamverse app, run one command to redeploy backend and frontend on GPU 4 with warmup enabled, then run the Playwright e2e suite against the live deployment.

Quick Start

Redeploy the Dreamverse app on GPU 4 with backend port 8009 and frontend port 5299 using the deploy script in this skill.

Frequently Asked Questions about dreamverse-deploy

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I redeploy the Dreamverse app on a specific GPU?▼

Run the dreamverse-deploy.sh script with the GPU index, backend port, and frontend port as positional arguments, for example GPU 4 with ports 8009 and 5299. The script tears down existing processes, exports the required environment, and waits for both services to return 200.

How do I enable torch.compile or GPU warmup for the Dreamverse backend?▼

Pass the --warmup and --torch-compile flags in any position relative to the positional arguments. Flags override the DREAMVERSE_WARMUP and DREAMVERSE_TORCH_COMPILE environment variables, and the readiness budget extends to 15 minutes when both are enabled.

Does the deploy script support NVENC hardware video encoding?▼

Yes, pass --nvenc to use h264_nvenc instead of libx264 software encoding. It requires a native ffmpeg built with NVENC support and a GPU with NVENC silicon; the script hard-fails during preflight if either is missing.

Why does a Dreamverse redeploy fail with GPU out of memory?▼

OOMs happen when a killed backend leaves orphaned multiproc_executor worker subprocesses holding GPU memory. The script handles this by nuking every process on the target GPU via nvidia-smi before launching, and waits up to 30 seconds for memory to release.

Can I use this deploy workflow for remote or container deployments?▼

No, this skill is scoped to a local checkout on a directly attached GPU. For containers use the Docker README under apps/dreamverse/docker, and for Modal follow the modal README; do not adapt the process-killing workflow to remote deployments.

Related Skills

Explore More Software Engineering Skills→
sayalinvidiasayalinvidia

cuopt-install

Guide cuOpt installation via pip, conda, or Docker with CUDA checks.

Community
Intermediate
skaile-aiskaile-ai

e2e-platform

Automate Skaile platform end-to-end tests with Playwright and auto-recovery.

Official
Advanced
yale-som-hpcyale-som-hpc
5

self-diagnosing-resource-use

Diagnose Slurm job CPU, memory, and GPU utilization using sacct, seff, and nvidia-smi.

Official
Intermediate
lightrain-alightrain-a

serverless-modal

Run GPU training and inference workloads on Modal with cost estimation.

Community
Advanced