Live control state
Overview
Local model catalog
Deploy a model
The deployment ends when its Slurm walltime elapses, so this is how long the model stays up.
Selected model details
Durable background import
Add Hugging Face Model
Profiles and import progress are shared with every Arena user.
Earlier imports
Slurm lifecycle
Deployments
| State | Profile | Deployment | Job | GPU | Expires | Action |
|---|
No deployments are known yet.
System details
Control path
- Bridge
- —Waiting for bridge
- Agent
- —No host data
- Allocated GPUs
- 0Policy loading
Command state
Scheduler
Versioned vLLM serving configuration
Recipes
Clone an Arena default, edit its vLLM arguments and non-secret container variables, then prepare an immutable revision for deployment.
Shared catalog
Serving recipes
| Name | Model | Image | Revision | Status | Deployments |
|---|
No serving recipes are available.
Command-style editor
Recipe detail
Select an Arena default or saved recipe.
Controlled generation
Inference playground
Live + durable result
Response
Idle
Run an inference request to view the response.
Thinking / reasoning
Prompt programs on the local host
Agents
Run the same workflow on several models and compare their responses side by side.
Model comparison
Responses
Idle
Run an agent to compare each model's response.
External model access
Inference endpoints
Publish a stable alias backed by a READY deployment. Traffic stays behind Arena's authenticated scheduler.
OpenAI-compatible API base
Stable model name
Create or repoint alias
Scoped credential
Create API key
Shown once
Copy this API key now
Arena stores only a keyed digest. This value cannot be recovered after you leave or create another key.
Quick start
Chat completions request
Supported now: one optional system message, exactly one user message, streaming or non-streaming output, bounded sampling fields, and compatible reasoning effort.
Published aliases
Endpoints and your keys
| Status | Alias | Deployment | Profile | Your API keys | Action |
|---|
No inference endpoint aliases have been published yet.
Persisted evaluation history
Runs
No runs match the current filter.
Selected record
Run detail
Select a run to inspect its prompt, response, usage, and metrics.