vastai-migration-deep-dive

Migrate GPU workloads to or from Vast.ai, or between GPU providers. Use when switching from AWS/GCP/Azure GPU instances to Vast.ai, migrating between GPU types, or re-platforming ML infrastructure. Trigger with phrases like "migrate to vastai", "vastai migration", "switch to vastai", "vastai from aws", "vastai from lambda".

Jump to install

Repository: jeremylongshore/tons-of-skills-marketplace
Last source activity: August 19, 2026 at 12:13
Detected SKILL.md language: English
Stars: 2,677
Forks: 392

Install options

The review-first prompt is selected by default. You can switch to a direct command or download a local copy.

Review the source files

Read SKILL.md and any companion files shown by SkillsMP before deciding whether to install.

Back to search Back to installation

name	vastai-migration-deep-dive
description	Migrate GPU workloads to or from Vast.ai, or between GPU providers. Use when switching from AWS/GCP/Azure GPU instances to Vast.ai, migrating between GPU types, or re-platforming ML infrastructure. Trigger with phrases like "migrate to vastai", "vastai migration", "switch to vastai", "vastai from aws", "vastai from lambda".
allowed-tools	Read, Write, Edit, Bash(vastai:), Bash(docker:), Grep
version	1.11.0
license	MIT
author	Jeremy Longshore <jeremy@intentsolutions.io>
tags	["saas","vast-ai","migration"]
compatibility	Designed for Claude Code

Vast.ai Migration Deep Dive

Current State

!vastai --version 2>/dev/null || echo 'vastai CLI not installed' !pip show vastai 2>/dev/null | grep Version || echo 'N/A'

Overview

Migrate GPU workloads to Vast.ai from hyperscaler providers (AWS, GCP, Azure) or other GPU clouds (Lambda, RunPod, CoreWeave). Also covers migrating between GPU types on Vast.ai and the reverse migration away from Vast.ai.

Prerequisites

Existing GPU workload with Docker image
Understanding of current GPU costs and utilization
Checkpoint-based training pipeline (for training migrations)

Instructions

Step 1: Cost Comparison Analysis

# Compare your current GPU costs against Vast.ai marketplace prices
PROVIDER_COSTS = {
    "aws_p4d.24xlarge":      {"gpu": "A100 40GB", "gpus": 8, "hourly": 32.77},
    "aws_p3.2xlarge":        {"gpu": "V100 16GB", "gpus": 1, "hourly": 3.06},
    "gcp_a2-highgpu-1g":     {"gpu": "A100 40GB", "gpus": 1, "hourly": 3.67},
    "azure_NC24ads_A100_v4": {"gpu": "A100 80GB", "gpus": 1, "hourly": 3.67},
    "lambda_1xA100":         {"gpu": "A100",      "gpus": 1, "hourly": 1.25},
}

VASTAI_TYPICAL = {
    "RTX_4090":  ,
    :      ,
    :  ,
}

 ():
    monthly_current = current_hourly *   
    monthly_vastai = vastai_hourly * 
    savings = monthly_current - monthly_vastai
    pct = (savings / monthly_current) * 
    ()
    ()
    ()

savings_analysis(, , , )

# Most Docker images work unchanged on Vast.ai
# Key differences:
# - Vast.ai instances run as root
# - /workspace is the default working directory
# - SSH access (not IAM roles) for authentication

# Adapt your existing Dockerfile
cat << 'DOCKERFILE' > Dockerfile.vastai
FROM your-existing-image:latest

# Vast.ai instances use /workspace by default
WORKDIR /workspace

# Install any Vast.ai-specific tools
RUN pip install boto3  # for S3 checkpoint uploads

# Copy training code
COPY src/ /workspace/src/
COPY configs/ /workspace/configs/

CMD ["python", "src/train.py"]
DOCKERFILE

docker build -t ghcr.io/org/training:vastai -f Dockerfile.vastai .
docker push ghcr.io/org/training:vastai

# On AWS/GCP: IAM roles provide automatic credentials
# On Vast.ai: Pass credentials explicitly via environment variables

# Create instance with env vars for cloud storage access
vastai create instance $OFFER_ID \
  --image ghcr.io/org/training:vastai \
  --disk 100 \
  --env "AWS_ACCESS_KEY_ID=AKIA... AWS_SECRET_ACCESS_KEY=... AWS_DEFAULT_REGION=us-east-1"

#!/bin/bash
set -euo pipefail
echo "Migration Validation Checklist"

# 1. Docker image runs on Vast.ai
vastai create instance $OFFER_ID --image ghcr.io/org/training:vastai --disk 50
# Wait for running...

# 2. GPU access works
ssh -p $PORT root@$HOST "nvidia-smi && python -c 'import torch; print(torch.cuda.is_available())'"

# 3. Cloud storage works
ssh -p $PORT root@$HOST "aws s3 ls s3://your-bucket/ | head -5"

# 4. Training runs and saves checkpoints
ssh -p $PORT root@$HOST "cd /workspace && python src/train.py --epochs 1 --checkpoint-dir /workspace/ckpt"

# 5. Checkpoints uploaded to cloud storage
ssh -p $PORT root@$HOST "aws s3 sync /workspace/ckpt/ s3://your-bucket/ckpt/"

# 6. Clean up
vastai destroy instance $INSTANCE_ID
echo "Migration validation complete"

## Rollback Procedure
1. Stop all Vast.ai instances: `vastai show instances` → `vastai destroy instance ID`
2. Re-provision on original cloud provider
3. Resume training from cloud-stored checkpoint
4. Vast.ai Docker image remains available for future retry

Factor	AWS/GCP/Azure	Vast.ai
Pricing	Fixed, premium	Variable, 50-90% cheaper
GPU availability	On-demand guaranteed	Marketplace (may sell out)
SLA	99.9% uptime	No SLA (spot instances)
IAM roles	Native	Manual credential passing
Networking	VPC, private subnets	Public SSH only
Storage	EBS/PD attached	Local disk + cloud storage
Support	Enterprise support	Community/email

Error	Cause	Solution
Docker image incompatible	Relies on IAM roles or cloud-specific APIs	Pass credentials via env vars
CUDA version mismatch	Different CUDA on Vast.ai hosts	Filter by `cuda_max_good` in search
Data transfer too slow	Large dataset over public internet	Stage data in cloud storage, download on instance
No matching offers	Specific GPU unavailable	Try alternative GPU type or wait for availability