It's the quickest way to stand up a model for tests, evals, or batch generation. (If you're after a managed, production-ready service instead, that's what Inference Endpoints are for — more on when to pick which at the end.)
Here's the whole thing end to end.
Prerequisites #
- A payment method or a positive prepaid credit balance (Jobs is billed per‑minute by hardware usage).
huggingface_hub >= 1.20.0
:pip install -U "huggingface_hub>=1.20.0"
.- Logged in locally:
hf auth login
.
Launch the server #
hf jobs run
is docker run
for HF infrastructure. We use the official vllm/vllm-openai
image, ask for a GPU with --flavor
, and expose vLLM's port with --expose
:
hf jobs run --flavor a10g-large --expose 8000 --timeout 2h \
vllm/vllm-openai:latest \
vllm serve Qwen/Qwen3-4B --host 0.0.0.0 --port 8000
--expose 8000
routes the container's port through HF's public jobs proxy (see the Serve Models guide for the full reference). The command prints the URL your server is reachable at:
✓ Job started
id: 6a381ca1953ed90bfb947332
url: https://huggingface.co/jobs/qgallouedec/6a381ca1953ed90bfb947332
Hint: Exposed ports are reachable at (requires an HF token with read access to the job):
https://6a381ca1953ed90bfb947332--8000.hf.jobs
6a381ca1953ed90bfb947332
is your job ID. Keep track of it, we'll need it. We'll use <job_id>
as a placeholder for it in the rest of the post.
Give it a couple of minutes to download weights and boot. When the logs show Application startup complete
, you're live.
Query it from anywhere #
vLLM speaks the OpenAI API, and every request just needs your HF token as a bearer token. The quickest way to hit it is curl:
curl https://<job_id>--8000.hf.jobs/v1/chat/completions \
-H "Authorization: Bearer $(hf auth token)" \
-H "Content-Type: application/json" \
-d '{
"model": "Qwen/Qwen3-4B",
"messages": [{"role": "user", "content": "Hello!"}],
"chat_template_kwargs": {"enable_thinking": false}
}'
which returns the usual OpenAI-style JSON, with choices[0].message.content
holding "Hello! How can I assist you today? 😊"
.
Or, from Python, point the OpenAI client at the exposed URL and pass the token as the API key:
from huggingface_hub import get_token
from openai import OpenAI
client = OpenAI(
base_url="https://<job_id>--8000.hf.jobs/v1",
api_key=get_token(),
)
resp = client.chat.completions.create(
model="Qwen/Qwen3-4B",
messages=[{"role": "user", "content": "Hello!"}],
extra_body={"chat_template_kwargs": {"enable_thinking": False}},
)
print(resp.choices[0].message.content)
Hello! How can I assist you today? 😊
Quick health check before you start: curl https://<job_id>--8000.hf.jobs/v1/models -H "Authorization: Bearer $(hf auth token)"
should list the model.
🔐 The endpoint is gated, not public.Every request must carry an HF token withread access to the job's namespace. A plain browser visit will be rejected. In effect, the jobs proxyisyour API gate: access is scoped to you (and your org). That's fine for private use, but treat the URL accordingly: don't share it expecting it to be open, and don't paste your token into untrusted places. If you need finer-grained or public access, put a proper gateway in front instead. Or see[HF Jobs or Inference Endpoints?]below.
Clean up #
Jobs are billed per second, so stop the server when you're done:
hf jobs cancel <job_id>
The --timeout
you set is a safety net (it'll auto-stop), but cancelling explicitly is cheaper. An a10g-large
runs at $1.50/hour — check hf jobs hardware
for the full price list and pick the smallest flavor that fits your model.
Going further: bigger models #
The same command scales to much larger models — pick a beefier --flavor
and tell vLLM to shard the model across the GPUs with --tensor-parallel-size
. For example, the 122B Qwen3.5 mixture-of-experts model on 2× H200:
hf jobs run --flavor h200x2 --expose 8000 --timeout 2h \
vllm/vllm-openai:latest \
vllm serve Qwen/Qwen3.5-122B-A10B \
--host 0.0.0.0 --port 8000 --tensor-parallel-size 2 \
--max-model-len 32768 --max-num-seqs 256
--tensor-parallel-size
should match the number of GPUs in the flavor (h200x2
→ 2, h200x8
→ 8). Run hf jobs hardware
to see what's available and give bigger models a longer --timeout
, since they take longer to download and load. For large models, H200 flavors are usually the best value.
The --max-model-len 32768 --max-num-seqs 256
flags are specific to this model: Qwen3.5-122B is a hybrid Mamba/attention architecture with a 256K-token default context, which doesn't leave enough memory for vLLM's default batch settings. Capping the context length and concurrent-sequence count keeps it within the GPUs' memory. If a model fails to start with an out-of-memory or cache-block error, dialing these two down is the first thing to try. Everything else (the exposed URL, the OpenAI client, the token auth) stays exactly the same.
Going further: Chat with it in a UI #
Prefer a chat window over curl? A few lines of Gradio point at the same endpoint. Add --reasoning-parser deepseek_r1
to the vllm serve
command so Qwen3's thinking comes back as a separate field (not necessary, but helpful), then run this code locally (you'll just need the job ID):
import gradio as gr
from gradio import ChatMessage
from huggingface_hub import get_token
from openai import OpenAI
client = OpenAI(base_url="https://<job_id>--8000.hf.jobs/v1", api_key=get_token())
def chat(message, history):
messages = [{"role": m["role"], "content": m["content"]} for m in history if not m.get("metadata")]
messages.append({"role": "user", "content": message})
stream = client.chat.completions.create(model="Qwen/Qwen3-4B", messages=messages, stream=True)
thinking, answer = "", ""
for chunk in stream:
delta = chunk.choices[0].delta
thinking += delta.model_extra.get("reasoning", "")
answer += delta.content or ""
out = []
if thinking.strip():
status = "done" if answer.strip() else "pending"
out.append(ChatMessage(role="assistant", content=thinking, metadata={"title": "💭 Thinking", "status": status}))
if answer.strip():
out.append(ChatMessage(role="assistant", content=answer))
yield out
gr.ChatInterface(chat).launch()
Run it, open http://127.0.0.1:7860
, and chat — reasoning streams into the collapsible panel, the answer below.
Going further: SSH into the running server #
Need to debug a startup failure, watch GPU memory, or tail logs interactively? You can open a shell straight into the running job. Launch it with --ssh
and make sure your public key is registered at huggingface.co/settings/keys:
hf jobs run --flavor a10g-large --expose 8000 --timeout 2h --ssh \
vllm/vllm-openai:latest \
vllm serve Qwen/Qwen3-4B --host 0.0.0.0 --port 8000
then connect with the job ID:
hf jobs ssh <job_id>
You're now inside the container, where you can run nvidia-smi
, inspect the process, or poke at the model directly — which makes debugging and monitoring much easier than reading logs from the outside. SSH support requires huggingface_hub >= 1.20.0
.
Going further: Use it as a coding-agent backend with Pi #
The same endpoint can back a terminal coding agent. Pi is a provider-agnostic agent harness. Point it at the job and you get a Read/Write/Edit/Bash agent running on your own self-hosted model.
One thing to set up first: agents drive the model through tool calls, and vLLM only accepts those if the server is launched with tool calling enabled. So relaunch with --enable-auto-tool-choice
and a --tool-call-parser
matching the model family (hermes
for Qwen3). Agents also benefit from a stronger model, so this is a good place to bring in the bigger one:
hf jobs run --flavor h200x2 --expose 8000 --timeout 2h \
vllm/vllm-openai:latest \
vllm serve Qwen/Qwen3.5-122B-A10B \
--host 0.0.0.0 --port 8000 --tensor-parallel-size 2 \
--max-model-len 32768 --max-num-seqs 256 \
--reasoning-parser deepseek_r1 \
--enable-auto-tool-choice --tool-call-parser hermes
Then add the job as a custom provider in ~/.pi/agent/models.json
:
{
"providers": {
"hf-jobs": {
"baseUrl": "https://<job_id>--8000.hf.jobs/v1",
"api": "openai-completions",
"apiKey": "!hf auth token",
"models": [
{ "id": "Qwen/Qwen3.5-122B-A10B" }
]
}
}
}
Then launch the agent against it:
pi
The model you spun up a couple of commands ago, now driving an interactive coding agent in your terminal.
HF Jobs or Inference Endpoints? #
HF Jobs isn't the only way to serve a model on Hugging Face. Inference Endpoints are our managed product for the same job, and which one fits depends on what you're after.
Reach for HF Jobs when you want maximum flexibility and control: it's just docker run
on HF infrastructure, so you pick the image, the exact vllm serve
flags, and the hardware, and you pay per second for as long as the job runs. That makes it a great fit for experiments, one-off evals, batch generation, or kicking the tires on a model before committing to anything.
Reach for Inference Endpoints when you want something more production-ready. They add the operational niceties a long-lived service needs: finer-grained access control (an endpoint can be public, protected, or private), and scale-to-zero, so you're not billed during periods of inactivity. If you're standing up a durable endpoint rather than running a job, that's the tool to grab.
Further reading #
This post sticks to vLLM, but the same expose-a-port pattern works with any OpenAI-compatible server. To serve GGUFs with llama.cpp or run SGLang instead, see the Serve Models on Jobs guide, which walks through those backends.