Single-node
Vast.ai runbook · model rank #8

Stand up Step 3.7 Flash
on Vast.ai.

Fits the provider's standard dedicated or managed endpoint path. Cost-sensitive dedicated containers when you can evaluate marketplace hosts yourself.

Open Vast.ai ↗ Read the runbook
Single-node: Fits the provider's standard dedicated or managed endpoint path.
Provider walkthrough

Vast.ai setup, in the order that matters

A Docker template attached to a matching GPU marketplace offer. Use the provider console for infrastructure and the pinned runtime block for the model server.

  1. 01

    Create a private template

    Use the vllm container in the runtime block below, expose port 8000, and keep HF_TOKEN in the template's secret environment.

  2. 02

    Filter the marketplace

    Find an offer with 1x H200-class GPU, at least 200GB of disk, reliable host scores, and enough internet bandwidth for a ~112GB checkpoint.

  3. 03

    Launch and inspect logs

    Launch from the private template, confirm every GPU is visible, and watch the model download and engine initialization before sending traffic.

  4. 04

    Expose only what you need

    Protect the OpenAI-compatible endpoint with your own gateway or tunnel. Stop or destroy the instance when the test is over.

Runtime seedvLLM · latest
docker run --rm --gpus all --ipc=host \
  -p 8000:8000 \
  -e HUGGING_FACE_HUB_TOKEN="$HF_TOKEN" \
  -v "$PWD/.hf-cache:/root/.cache/huggingface" \
  vllm/vllm-openai:latest \
  --model stepfun-ai/Step-3.7-Flash \
  --tensor-parallel-size 1 \
  --max-model-len 32768 \
  --trust-remote-code \
  --served-model-name step-3.7-flash

This is a reproducible starting block. Cluster paths still need the model author's distributed recipe and the provider's orchestration layer.

Smoke testOpenAI-compatible request
curl http://127.0.0.1:8000/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "step-3.7-flash",
    "messages": [{"role": "user", "content": "Reply with: deployment healthy"}],
    "max_tokens": 32
  }'

Run this only after logs show the model loaded. A successful HTTP response is not a latency or quality benchmark.

Before you launch

Quota, license, and storage

  • Confirm 1x H200-class GPU is actually available in the selected region.
  • Read the Apache 2.0 terms and accept any gated-model conditions.
  • Budget at least 200GB for weights, cache, and container layers.
  • Keep HF_TOKEN in the provider secret store—not in scripts or templates.
Before real traffic

Memory, format, and shutdown

  • Record idle/free VRAM after the model loads and after a representative prompt.
  • Validate the official chat template, reasoning parser, and tool-call parser.
  • Add authentication and TLS in front of port 8000.
  • Verify the provider's stop/delete action actually ends compute billing.
Agent-assisted setup

Hand this runbook to Claude Code or Codex

Open a terminal in the repository where you want the deployment files, start claude or codex, then paste this prompt. It asks the agent to verify sources and stop before it creates billable infrastructure.

Claude Code quickstart ↗ Codex docs ↗
Deployment prompt 
Deploy Step 3.7 Flash (stepfun-ai/Step-3.7-Flash) on Vast.ai. Use this guide as the starting context: https://www.getflops.ai/models/step-3.7-flash/vast-ai. Use a planning floor of 1x 120GB GPUs, 200GB storage, vllm latest, and an initial context limit of 32768 tokens. Open every linked primary source and flag any mismatch instead of guessing. Create a deployment folder containing README.md, .env.example with no secrets, a pinned start script or infrastructure manifest, and smoke-test.sh. Make the endpoint OpenAI-compatible where the runtime supports it. Run local/static validation, estimate the billable resources, and stop before provisioning paid infrastructure until I approve.
Guardrails included No secrets in files · verify primary docs · approval before spend
Research trail

The two sources to check first

StepFun model card Model size, context, license, chat format, and engine notes ↗ Vast.ai template docs A Docker template attached to a matching GPU marketplace offer ↗

Checked 2026-07-27. This page separates sourced facts from the planning baseline; verify current runtime support and capacity before spending.

Same model, another cloud

Compare Step 3.7 Flash elsewhere

RunPod Guided endpoint 1x H200-class GPU Lambda Single-node 1x H200-class GPU AWS Single-node SageMaker/HyperPod GPU capacity with at least 1x H200-class accelerators Google Cloud Single-node Vertex AI GPU deployment with at least 1x H200-class accelerators Azure Single-node Azure ML GPU deployment with at least 1x H200-class accelerators Oracle Cloud Single-node OCI GPU shape with at least 1x H200-class accelerators